Information processing device, control system, search method, and search program

Abstract

To provide an information processing device capable of stably performing feedback control free of the subjective view of a person.SOLUTION: An information processing device comprises: a predictive distribution calculation unit (102) which calculates a prediction distribution of a function indicative of a relation between a control parameter in feedback control and results of feedback control applied with the control parameter; and a search unit (103) which searches for an optimum value candidate of the control parameter on the basis of the prediction distribution.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] This invention relates to a technique for adjusting control parameters in feedback control. [Background technology]

[0002] Feedback control is a control method that controls input values ​​to match output values ​​with target values, and it has long been used for the automatic control of various devices. To perform proper feedback control, it is necessary to set the control parameters appropriately, and techniques for appropriately setting these parameters have been known for a long time.

[0003] For example, Patent Document 1 discloses an exhaust gas denitrification control device comprising an injection amount feedback control means for calculating the amount of material injected to remove nitrogen oxides from combustion exhaust gas discharged from a combustion furnace, and a parameter adjustment means for adjusting the control parameters of the feedback control based on the furnace opening temperature of the combustion furnace. [Prior art documents] [Patent Documents]

[0004] [Patent Document 1] Special Publication No. 7-71619 [Overview of the project] [Problems that the invention aims to solve]

[0005] The conventional techniques described above assume a correlation between furnace port temperature and denitrification rate, and between furnace port temperature and process delay, and determine appropriate control parameters for feedback control of injection volume from the furnace port temperature. However, these techniques cannot be applied to anything other than feedback control of injection volume. Thus, no method for adjusting control parameters that is commonly applicable to various feedback controls has been known until now. For this reason, the control parameters for feedback control have often been determined based on the operator's experience and intuition.

[0006] However, relying on experience and intuition to determine control parameters results in inconsistencies depending on who makes the determination, making it difficult to achieve stable feedback control.

[0007] One aspect of the present invention aims to provide an information processing device, etc., that enables stable feedback control without being influenced by human subjectivity. [Means for solving the problem]

[0008] To solve the above problems, an information processing device according to one aspect of the present invention comprises a predictive distribution calculation unit that calculates a predictive distribution of a function showing the relationship between control parameters in feedback control and the result of feedback control to which the control parameters are applied, and a search unit that searches for candidate optimal values ​​of the control parameters based on the predictive distribution.

[0009] Furthermore, a control system according to one aspect of the present invention includes, in order to solve the above problems, a control device that performs feedback control on a device to be controlled, and an information processing device that searches for candidate optimal values ​​of the control parameters based on a predictive distribution of a function showing the relationship between the control parameters in the feedback control and the result of the feedback control to which the control parameters are applied, wherein the control device performs feedback control on the device to be controlled by applying the candidate.

[0010] Furthermore, a search method according to one aspect of the present invention is a search method performed by one or more information processing devices in order to solve the above problems, and includes a predictive distribution calculation step of calculating a predictive distribution of a function that shows the relationship between control parameters in feedback control and the results of feedback control to which the control parameters are applied, and a search step of searching for candidate optimal values ​​of the control parameters based on the predictive distribution. [Effects of the Invention]

[0011] According to one aspect of the present invention, it becomes possible to perform stable feedback control without being influenced by human subjectivity.

Brief Description of the Drawings

[0012] [Figure 1] It is a block diagram showing an example of the main configuration of an information processing apparatus according to an embodiment of the present invention. [Figure 2] It is a diagram showing an overview of a control system including the above information processing apparatus. [Figure 3] It is a diagram showing an example of a display screen for receiving input of result data. [Figure 4] It is a diagram showing an example of a display screen for search results. [Figure 5] It is a diagram showing a histogram of measured values (PV) measured during a period in which feedback control is performed. <000008​​​​​​​​​​​​​​​​​​The user of the control system 9 can perform stable feedback control independent of the user's subjectivity by applying the candidates detected by the information processing apparatus 1 to the feedback control in the control system 9.

[0016] As described above, the control device 3 is a device that performs feedback control on the device 4 to be controlled. Specifically, the control device 3 performs feedback by PID (Proportional Integral Differential) control. In PID control, the manipulated variable (MV) is determined by three elements: the deviation (DV) between the set value (SV) and the measured value (PV), the integral of DV, and the differential of DV. The calculation formula for the manipulated variable (MV) in PID control is expressed as follows.

[0017] MV = P×DV+(60 / I)×(integral of DV)+(D / 60)×(differential of DV) In the above calculation formula, P, I, and D are control parameters that determine the degree of emphasis on proportional control, integral control, and differential control, respectively. In order to achieve appropriate feedback control, it is necessary to appropriately set the values of P, I, and D, and the information processing apparatus 1 searches for the optimal values of these control parameters. Note that the information processing apparatus 1 does not necessarily have to search for the optimal values of all of the control parameters of P, I, and D, and it may search for at least one of them.

[0018] In the control system 9, the operation terminal to which the manipulated variable (MV) calculated by the above calculation formula is applied is the device 4 or its component. For example, when the device 4 includes a valve, the valve may be used as the operation terminal, and the opening degree of the valve may be used as the manipulated variable (MV). Also, for example, when the device 4 includes an inverter, the inverter may be used as the operation terminal.

[0019] If a single device 4 has multiple control terminals, the control device 3 determines the manipulated amount (MV) for each control terminal. In this case, the information processing device 1 may detect candidate optimal values ​​for the control parameters for each control terminal. For example, when controlling the temperature of device 4, the manipulated amount (MV) is calculated for each of the two control terminals, cooling and heating, using the calculation formula described above. In this case, the information processing device 1 can simultaneously detect candidate optimal values ​​for the control parameters used for cooling control and candidate optimal values ​​for the control parameters used for heating control. Of course, the information processing device 1 can also detect candidate optimal values ​​for the control parameters used for heating control after detecting candidate optimal values ​​for the control parameters used for cooling control. Similarly, the information processing device 1 can also detect candidate optimal values ​​for the control parameters used for cooling control after detecting candidate optimal values ​​for the control parameters used for heating control.

[0020] Furthermore, even when targeting the same operating terminal, different control parameters may be applied depending on the conditions. In this case, the information processing device 1 may detect candidate optimal values ​​for the control parameters for each condition. For example, when the device 4 is operating under the first operating conditions, the control device 3 may control the device 4 with an manipulated variable (MV) calculated using the first control parameter, while when the device 4 is operating under the second operating conditions, it may control the device 4 with an manipulated variable (MV) calculated using the second control parameter. In this case, the information processing device 1 can detect candidate optimal values ​​for both the first and second control parameters at the same time. Of course, the information processing device 1 can also detect candidate optimal values ​​for the second control parameter after detecting candidate optimal values ​​for the first control parameter. Similarly, the information processing device 1 can also detect candidate optimal values ​​for the first control parameter after detecting candidate optimal values ​​for the second control parameter.

[0021] The plant processes in control system 9 include, for example, combustion and chemical reactions. However, control system 9 is not limited to plant processes and can be applied to the control of any object.

[0022] The measurement terminal in the control system 9 is for measuring the measured value (PV), and is implemented, for example, by a measuring device 5. The measuring device 5 should be appropriate for the object being measured. For example, if the measured value (PV) is pressure, a pressure gauge should be used as the measuring device 5, and if the measured value (PV) is temperature, a thermometer should be used as the measuring device 5. The measured value (PV) measured by the measuring device 5 is used to calculate the deviation (DV) as described above, and is also stored in the database 2 and used by the information processing device 1 to search for control parameters. Note that the database 2 may be provided by the information processing device 1 or the control device 3.

[0023] As described above, the control system 9 includes a control device 3 that performs feedback control on the controlled device 4, and an information processing device 1 that searches for candidate optimal values ​​for the control parameters based on a predictive distribution of a function showing the relationship between the control parameters in the feedback control and the results of the feedback control to which the control parameters are applied. The control device 3 applies the candidates detected by the information processing device 1 to perform feedback control on the controlled device 4. This control system 9 makes it possible to perform stable feedback control without being influenced by human subjectivity. In addition, it reduces the time required to consider control parameters, thus contributing to the efficiency of the management operations of the control system 9.

[0024] The feedback control targeted by the information processing device 1 for detecting candidate optimal values ​​for control parameters may be, for example, feedback control in a waste incineration facility. Generally, waste incineration facilities incinerate a wide variety of waste, and the properties of this waste (e.g., moisture content) are not constant, so the operating conditions are constantly changing. For this reason, feedback control that responds to the operating conditions is required, but it has been difficult to perform highly accurate control due to factors such as the instability of the properties of the waste. In this respect, the information processing device 1 makes it possible to perform stable feedback control by searching for candidate optimal values ​​for control parameters based on a predictive distribution.

[0025] Furthermore, if the control system 9 performs feedback control of the waste incineration facility, then equipment 4 will be equipment located in the waste incineration facility. For example, equipment 4 may be an incinerator, a grate for transporting waste within the incinerator, a combustion air supply device, or an exhaust gas treatment device. In addition, the waste incineration facility may be equipped with a power generation facility that generates electricity using the waste heat from the incinerator, in which case equipment 4 may be equipment related to the power generation facility (e.g., a boiler).

[0026] [Configuration of the information processing device] The configuration of the information processing device 1 will be explained based on Figure 1. Figure 1 is a block diagram showing an example of the main components of the information processing device 1. As shown in the figure, the information processing device 1 includes a control unit 10 that controls all parts of the information processing device 1, and a storage unit 11 that stores various data used by the information processing device 1. The information processing device 1 also includes a communication unit 12 for the information processing device 1 to communicate with other devices, an input unit 13 that receives input of various data to the information processing device 1, and an output unit 14 for the information processing device 1 to output various data. The control unit 10 also includes a data acquisition unit 101, a prediction distribution calculation unit 102, a search unit 103, a reception unit 104, a prediction value presentation unit 105, a reason presentation unit 106, a change pattern presentation unit 107, and an update necessity determination unit 108.

[0027] The data acquisition unit 101 acquires result data that associates the control parameters in the feedback control with the results of the feedback control to which those control parameters are applied. Furthermore, after the search unit 103 has detected candidate optimal values ​​for the control parameters, the data acquisition unit 101 acquires result data that shows the results of the feedback control to which those candidates are applied.

[0028] The predictive distribution calculation unit 102 calculates a predictive distribution of a function that shows the relationship between the control parameters in feedback control and the results of feedback control to which those control parameters are applied. Furthermore, after calculating the predictive distribution, if the data acquisition unit 101 acquires new result data (showing the results of feedback control to which the candidates detected by the search unit 103 are applied), the predictive distribution calculation unit 102 updates the predictive distribution using the new result data.

[0029] While acquiring result data and updating the predictive distribution are not essential steps in this configuration, this configuration allows for the detection of more valid candidate optimal values ​​for control parameters based on the updated predictive distribution. Furthermore, the acquisition of result data and updating of the predictive distribution can be repeated, enabling high-precision feedback control by selecting candidate control parameters as the optimal value or a value close to the optimal value.

[0030] The search unit 103 searches for candidate optimal values ​​for the control parameters based on the predictive distribution calculated by the predictive distribution calculation unit 102. Details of the method for calculating and updating the predictive distribution and the method for searching for candidates will be described later.

[0031] The reception unit 104 accepts a selection from the search unit 103 regarding whether to prioritize the search for candidates around known values ​​or within the unknown range of feedback control results. User input for this selection can be received, for example, via the input unit 13. When the reception unit 104 accepts the above selection, the search unit 103 searches for candidates according to that selection.

[0032] The reception unit 104 is not a mandatory component, but if the information processing device 1 is equipped with the reception unit 104, the user of the information processing device 1 can choose to search by emphasizing the vicinity of known values ​​for the feedback control results. In this case, it is possible to detect candidates that may be local optima but are expected to provide stable control results. Alternatively, the user of the information processing device 1 can choose to search by emphasizing the unknown range for the feedback control results. In this case, there is a possibility of hits and misses in the control results, but it is possible to detect candidates that are expected to contribute significantly to the detection of a global optimum.

[0033] The prediction value presentation unit 105 presents the predicted value resulting from feedback control applied to the candidates detected by the search unit 103. As will be described in detail later, the prediction value presentation unit 105 calculates the above predicted value from the mean and variance of the prediction distribution.

[0034] The predicted value presentation unit 105 is not an essential component, but if the information processing device 1 is equipped with the predicted value presentation unit 105, it can present the user with useful information to help them decide whether or not to apply the detected candidate, namely the predicted value of the feedback control result to which the detected candidate has been applied.

[0035] The reasoning unit 106 presents the reasons why the candidate detected by the search unit 103 is valid. As will be described in detail later, the reasoning unit 106 identifies the above reasons based on the mean and variance of the predictive distribution.

[0036] While the reason-presenting unit 106 is not an essential component, if the information processing device 1 is equipped with the reason-presenting unit 106, it can provide the user with useful information to help them decide whether or not to apply a detected candidate, namely the reasons why that candidate is valid.

[0037] The change mode presentation unit 107 presents information indicating how the manipulated variable (MV) for the controlled device will change when the candidate detected by the search unit 103 is applied. As will be described in detail later, the change mode presentation unit 107 identifies how the manipulated variable (MV) will change based on the difference between the values ​​of the control parameters to be compared (coefficients P, I, and D in the calculation formula described above) and the candidate.

[0038] Although the change pattern presentation unit 107 is not an essential component, if the information processing device 1 is equipped with the change pattern presentation unit 107, it can present the user with useful information to help them decide whether or not to apply the detected candidate, such as information showing how the operation amount changes when the candidate is applied.

[0039] The update necessity determination unit 108 acquires the results of feedback control applied to the candidates detected by the search unit 103, and repeatedly performs a process at predetermined intervals to determine whether or not to update the predicted distribution based on the acquired results.

[0040] The update necessity determination unit 108 is not an essential component, but if the information processing device 1 is equipped with the update necessity determination unit 108, it can automatically continue updating the predicted distribution and detecting candidates until the result of the feedback control becomes desired, and can automatically terminate updating the predicted distribution and detecting candidates when the result of the feedback control becomes desired. Furthermore, if the information processing device 1 is equipped with the update necessity determination unit 108, it can automatically resume updating the predicted distribution and detecting candidates after the result of the feedback control becomes desired, if the result is no longer desired.

[0041] Furthermore, the update necessity determination unit 108 may also determine whether there are any problems with the operation of the device 4 after the candidate detected by the search unit 103 is applied, and if it determines that there are problems, it may perform a process to change the control parameters. This minimizes the impact if the detected candidate is not appropriate. Note that these processes may also be performed by a separate processing block from the update necessity determination unit 108.

[0042] In feedback control, if we can find a function that shows the relationship between the control parameters and the results of the feedback control to which those control parameters are applied, it will be easy to determine the control parameters that will produce the desired result. However, in general, it is difficult to find such a function.

[0043] Therefore, instead of finding a function that shows the relationship between the control parameters and the control result, the information processing device 1 employs a configuration in which it calculates the predictive distribution of the said function and searches for candidate optimal values ​​for the control parameters based on the calculated predictive distribution.

[0044] This allows for the automatic determination of appropriate control parameters from a predictive distribution, even when it is difficult to determine a function that shows the relationship between control parameters and control results. By applying these determined control parameters, stable feedback control becomes possible, independent of human subjectivity.

[0045] [Details on the method for calculating the predictive distribution and the method for searching for candidate optimal values] The details of the method for calculating (including updating) the predictive distribution by the predictive distribution calculation unit 102 and the details of the method for searching for candidate optimal values ​​by the search unit 103 are described below. Note that the following explanation pertains to the method for calculating and searching for the predictive distribution in Bayesian optimization. However, other methods besides those described below can be applied as long as they use the predictive distribution to search for candidate optimal values.

[0046] In feedback control, if there are N control parameters to be optimized, then those control parameters are:

[0047]

number

[0048] It is expressed as, The evaluation value for that is,

[0049]

number

[0050] It is expressed as follows.

[0051] The predictive distribution calculation unit 102 uses the result data acquired by the data acquisition unit 101 to calculate the predictive distribution of a function that shows the relationship between control parameters and control results. This function will be referred to as the evaluation function f(θ) below. When new result data (for example, result data showing the result of feedback control to which a candidate detected by the search unit 103 has been applied) is acquired, the predictive distribution calculation unit 102 updates the predictive distribution to reflect this new result data.

[0052] The relationship between the control variable and the control result is expressed using Gaussian noise ε. n Using ~N(0,β)

[0053]

number

[0054] Assuming this, the following distribution is obtained as the predictive distribution of the evaluation function by the Gaussian process.

[0055]

number

[0056]

number

[0057]

number

[0058] Here, k * =k(θ,θ), and K Θ is [KΘ ) i,j = k(θ i , θ j ) is the Gram matrix obtained. Also,

[0059]

Number

[0060] is. k Θ,* is the vertical vector of [k Θ,* i = k(θ i , θ), and k(·, ·) is the kernel function. Here, the parameter of the kernel function is α k is set as.

[0061] The mean function μ(θ) represents the average value of the evaluation function predicted from the result data. Also, the variance function σ(θ) is the variance of the evaluation function predicted from the result data. σ(θ) indicates the uncertainty of the prediction, and its value tends to increase in regions where the result data is insufficient. When σ is large, it can be said that the prediction is uncertain and that there is a lack of result data necessary to increase the certainty of the prediction. As is clear from Equation (3), the kernel function and the kernel function parameter α k included in the variance function σ(θ) affect the calculation of the prediction distribution. When calculating the prediction distribution, the parameter α k is optimized. The optimization method is not particularly limited, and for example, various optimization techniques applied in general Bayesian optimization can also be applied.

[0062] The search unit 103 searches for candidates for the optimal control parameter in order to obtain the optimal control parameter. Specifically, as shown in the following Equations (4) and (5), the search unit 103 searches for the control parameter that maximizes the acquisition function a(θ) using the mean function μ(θ) and the variance function σ(θ). The control parameter detected in this search becomes a candidate for the optimal control parameter. This search is based on the UCB (Upper Confidence Bound) strategy. [[ID=^3]]

[0063] ​ In equation (5), κ is a parameter for adjusting the search and utilization. When the reception unit 104 receives a selection of whether to prioritize the search around a known value or the unknown range based on the feedback control result, the search unit 103 sets κ to a value corresponding to the received selection and performs a search for candidates.

[0064] Of course, it is also possible to explore new control parameters using other methods. For example, one could explore candidate optimal control parameters using a Probability of Improvement (PI) strategy or an Expected Improvement (EI) strategy.

[0065] Furthermore, when seeking the optimal control parameters that minimize the value of the evaluation function (for example, when the time required to complete a task is used as the evaluation value for a task that should ideally be completed in a short amount of time), one should search for the control parameters that minimize the acquisition function a(θ).

[0066]

number

[0067]

number

[0068] The search unit 103 detects candidate optimal values ​​for control parameters, which are then presented to the user. Feedback control is then initiated using these candidates, and the measured values ​​(PV) taken during the period in which the control is performed are stored in database 2.

[0069] The update necessity determination unit 108 obtains measurement values ​​(PV) from the database 2 that indicate the result of feedback control applied to the candidate detected by the search unit 103, and determines whether or not to update the predicted distribution based on the obtained measurement values ​​(PV). For example, the update necessity determination unit 108 may use the obtained measurement values ​​(PV) to calculate an evaluation value indicating the validity of the applied candidate, and if the calculated evaluation value is less than a predetermined threshold, it may determine to update the predicted distribution.

[0070] Furthermore, if the update necessity determination unit 108 determines that the predictive distribution should be updated, it may generate result data by associating the calculated evaluation value with the acquired measurement value. The data acquisition unit 101 acquires the generated result data and outputs it to the predictive distribution calculation unit 102 along with the previously acquired result data. The predictive distribution calculation unit 102 then updates the predictive distribution using the newly acquired result data and the previously acquired result data. Note that the calculation of the evaluation value may be performed by the update necessity determination unit 108 or the data acquisition unit 101, or by another device or user. The method for calculating the evaluation value will be described later.

[0071] The search unit 103 then searches for the optimal values ​​of the control parameters based on an evaluation function constructed using the updated predicted distribution calculated by the predicted distribution calculation unit 102. By repeatedly updating the predicted distribution and searching for the optimal values ​​in this way, it becomes possible to detect the optimal control parameters.

[0072] [Example of display screen (input / settings)] The data acquisition unit 101 may display a screen like the one shown in Figure 3 to allow the user to input the result data of the feedback control. Figure 3 is a diagram showing an example of a display screen that accepts the input of result data. As will be explained below, by displaying such a screen, the user can easily input the result data necessary for optimization and make various settings.

[0073] The display screen A1 shown in Figure 3 includes an address input box A11, a reference button A12, a confirmation button A13, and a result data display area A14. The address input box A11 is for entering the address of a file containing the result data. The reference button A12 is for referencing a file containing the result data. The user can specify the file to use by entering the address of the file containing the result data into the address input box A11, or by operating the reference button A12 and specifying the reference destination. For example, it may be possible to specify a CSV (Comma Separated Variables) format file, in which case the data acquisition unit 101 should read the result data from the specified CSV format file.

[0074] After specifying a file, the user can confirm the file to be used by operating the confirmation button A13. As a result, the result data read from the confirmed file will be displayed in the result data display area A14. In other words, the data acquisition unit 101 reads result data (control parameter values ​​and their evaluation values) from the specified file and displays these values ​​in the result data display area A14. In the example in Figure 3, the values ​​of the control parameters P, I, and D are displayed in correspondence with the evaluation values ​​calculated from the measured values ​​(PV) measured during the period in which the feedback control to which those values ​​were applied was applied.

[0075] The destination of display screen A1 is not particularly limited. For example, the data acquisition unit 101 may display display screen A1 on the output unit 14 if the output unit 14 is a display device, or it may display display screen A1 on another display device that is wired or wirelessly connected to the information processing device 1. Furthermore, the processing to display display screen A1 may be performed by the reception unit 104, or a separate processing block may be provided to perform the processing to display the display screen, and this processing block may perform the processing.

[0076] Furthermore, display screen A1 also contains various information for accepting optimization settings. Specifically, display screen A1 includes a manufacturer selection box A15, a calculation parameter selection box A16, an affirmative button A17, and a negative button A18.

[0077] Manufacturer specification box A15 is for specifying the manufacturer of the control system 9 (more specifically, the manufacturer of the DCS: Distributed Control System used in the control system 9). While the way PID control parameters are expressed may differ depending on the manufacturer, accepting manufacturer specification allows for optimization of the control parameters expressed in a manner specific to that manufacturer.

[0078] The calculation parameter specification box A16 accepts the selection of whether to prioritize searching for candidate optimal values ​​for control parameters around known values ​​based on feedback control results, or within the range of unknown feedback control results.

[0079] Specifically, the calculation parameter specification box A16 allows the user to enter one of three values ​​from 1 to 3. A value of "1" indicates that the system prioritizes utilization, meaning that the feedback control results prioritize exploring around known values. A value of "3" indicates that the feedback control results prioritize exploring in the unknown range. A value of "2" indicates that the system performs a standard search that is not biased towards either utilization or exploration. This allows the user to perform the search in the desired manner. Of course, the method of specifying the search manner is not limited to this example; for example, the user could be allowed to specify the value of κ in the above formula (5).

[0080] The user verifies whether the result data displayed in the result data display area A14 on the display screen A1 is valid. The user also verifies whether the inputs in the manufacturer specification box A15 and the calculation parameter specification box A16 in the "Settings" section are valid. If the user determines that both are valid, they press the affirmative button A17. This starts the calculation of the predicted distribution by the predicted distribution calculation unit 102, followed by the search unit 103 searching for candidate optimal values ​​for the control parameters. If the user wishes to cancel the optimization, they can press the negative button A18.

[0081] [Display screen example (result display)] The search unit 103 may display a screen like the one in Figure 4 to present the search results to the user. Figure 4 is a diagram showing an example of a search result display screen. The display screen A2 shown in Figure 4 includes a candidate display area A21 that displays candidate optimal values ​​for the control parameters detected by the search unit 103. In the example in Figure 4, the values ​​of three control parameters, P, I, and D, are displayed in the candidate display area A21.

[0082] Furthermore, display screen A2 includes a "Reference Information" section. This section contains three items (A22-A24): the expected evaluation value, the reason for the proposal, and the pattern of change in the manipulated amount. Displaying the "Reference Information" is not mandatory, but it is preferable to display it as it provides information for deciding whether or not to apply the candidates shown in candidate display area A21.

[0083] The predicted evaluation value item A22 displays the predicted evaluation value calculated by the prediction value presentation unit 105 when the control parameters shown in the candidate display area A21 are applied. Specifically, the prediction value presentation unit 105 identifies the mean and variance of the evaluation value when the control parameters detected as candidates are applied from the latest prediction distribution calculated by the prediction distribution calculation unit 102, and calculates the value of "mean ± standard deviation (square root of variance)" as the predicted evaluation value. For example, if the mean is 83 and the standard deviation is 12, the predicted evaluation value will be "83 ± 12" as shown in the figure. Note that the predicted value expressed using the standard deviation is just one example of reference information, and various other information can be applied as reference information. For example, the calculated variance value may be displayed as reference information.

[0084] Item A23 of the reason for the proposal presents the reasons why it is preferable to apply the control parameters shown in the candidate display area A21, as identified by the reason presentation unit 106. Specifically, item A23 displays the message "The certainty is low, but a high score can be expected." The content of the message can be determined based on the mean and variance mentioned above. For example, the reason presentation unit 106 may determine the message to present according to the rule that if the mean is above a predetermined threshold, it presents a message indicating that a high score can be expected; if the variance is above a predetermined threshold, it presents a message indicating that the certainty is low; and if it is below the threshold, it presents a message indicating that the certainty is high. In this case, if both the mean and variance are above the threshold, a message like the one shown in Figure 4 will be presented. On the other hand, if the mean is above the threshold but the variance is below the threshold, the reason presentation unit 106 may present a message indicating that the certainty is high and a high score can be expected. Furthermore, if both the mean and variance are below the threshold, the reason presentation unit 106 may present a message indicating that a high score cannot be expected, but the certainty is high. Furthermore, if the mean is below the threshold and the variance is above the threshold, the reason presentation unit 106 may display that there is no reason for the proposal, or that it is desirable to add result data.

[0085] Item A24, which describes the change in the manipulated variable, shows the change in the manipulated variable (MV in Figure 2) before and after applying the control parameter shown in the candidate display area A21, as identified by the change pattern presentation unit 107. Specifically, messages are displayed indicating that the manipulated variable (MV) with respect to the deviation (DV) increases, the correction amount of the manipulated variable (MV) required to make the deviation (DV) zero increases, and the rate of change of the manipulated variable (MV) slows down.

[0086] It is assumed here that the control parameters include the values ​​of the coefficients P, I, and D in the formula for calculating the manipulated variable (MV) described above. Furthermore, it is assumed that the feedback control calculates the manipulated variable (MV) for the controlled device from the deviation (DV) between the setpoint (SV) and the measured value (PV), and the coefficients mentioned above, as shown in the formula.

[0087] For example, if the three coefficients P, I, and D described above are optimized, the change mode presentation unit 107 may display a message indicating that the manipulated variable (MV) will increase if P increases before and after applying the control parameters shown in the candidate display area A21, as shown in Figure 4. On the other hand, if P decreases, the change mode presentation unit 107 may display a message indicating that the manipulated variable (MV) will decrease.

[0088] Furthermore, if I decreases before and after applying the control parameters shown in the candidate display area A21, the change mode presentation unit 107 may present a message indicating that the correction amount of the manipulated variable (MV) required to make the deviation (DV) zero will increase, as shown in Figure 4. On the other hand, if I increases, the change mode presentation unit 107 may present a message indicating that the correction amount will decrease.

[0089] Furthermore, if D decreases before and after applying the control parameter shown in the candidate display area A21, the change mode presentation unit 107 may display a message indicating that the rate of change of the manipulated variable (MV) will slow down, as shown in Figure 4. On the other hand, if D increases, the change mode presentation unit 107 may display a message indicating that the rate of change will speed up.

[0090] After confirming the above information, the user can simply press the finish button A25 to terminate the optimization process. Following this, the control parameters shown in the candidate display area A21 are transmitted to the control device 3, and feedback control is performed applying these parameters.

[0091] [Regarding evaluation values] The evaluation value of the feedback control result should be appropriately chosen according to the target and purpose of the feedback control. For example, if it is desirable to stabilize the measured value (PV), an evaluation value indicating the percentage of measured values ​​(PV) within a predetermined tolerance range among the measured values ​​(PV) measured during the period in which the feedback control was performed may be used as information indicating the result of the feedback control. This makes it possible to detect candidate control parameters that increase the percentage of measured values ​​within the tolerance range and stabilize the measured value (PV) by applying these to the feedback control.

[0092] The calculation of the evaluation value may be performed by the data acquisition unit 101 or the update necessity determination unit 108. Alternatively, a separate processing block for calculating the evaluation value may be provided, or the evaluation value (or result data including the evaluation value) calculated by another device or user may be input to the information processing device 1.

[0093] Furthermore, the predetermined tolerance range may be defined as the range from the value obtained by subtracting the standard deviation of the measured values ​​(PV) measured during the period in which feedback control was performed from the mean value of the measured values ​​(PV) measured during that period, to the value obtained by adding the standard deviation to the mean value. In other words, the range of "mean value ± standard deviation" may be defined as the predetermined tolerance range. This allows for a more stable evaluation compared to, for example, using the variance of the measured values ​​(PV) measured during that period as the evaluation value.

[0094] This will be explained using Figure 5. Figure 5 is a histogram of the measured values ​​(PV) measured during the period in which feedback control was performed. The histogram H shown in Figure 5 divides the measured values ​​(PV) into 10 classes according to their values, and shows the frequency of each class. In histogram H, the frequency is high for classes near the center and decreases for classes further from the center.

[0095] As shown in the figure, in histogram H, the range of measured values ​​(PV) from the value obtained by subtracting the standard deviation of a measured value (PV) from the mean value to the value obtained by adding the standard deviation to the mean value, i.e., the range of "mean ± standard deviation", is included in the four classes in the middle. The higher the proportion of measured values ​​(PV) that fall into these classes out of all measured values ​​(PV), the more stable the measured values ​​(PV) are considered to be. Therefore, the evaluation value calculated by the following formula can be said to be appropriate for evaluating stability.

[0096] (Evaluation value) = {Number of measured values ​​(PV) within the range of mean ± standard deviation / Total number of measured values ​​(PV)} × 100 Furthermore, since the evaluation value calculated using the above formula is based on the average value, it will show a reasonable evaluation result without being affected even if the average value rises or falls due to factors such as the manipulated variable (MV) being too large or too small. Therefore, as described above, a stable evaluation can be performed by applying this evaluation value.

[0097] Of course, the evaluation value can be anything appropriate to the target and purpose of the feedback control, and is not limited to the examples above. For example, the variance of the measured value (PV) could be used as the evaluation value.

[0098] [Processing flow] The flow of processing (search method) executed by the information processing device 1 will be explained based on Figure 6. Figure 6 is a flowchart showing an example of processing executed by the information processing device 1.

[0099] In S1, the data acquisition unit 101 acquires the result data of the feedback control. For example, the data acquisition unit 101 may display a screen as shown in Figure 3 and have the user input the result data of the feedback control. At this time, the reception unit 104 may also accept the optimization settings. Specific settings may include, for example, specifying the manufacturer of the control system, or selecting whether to prioritize the search between the area around known values ​​for the feedback control results and the unknown range for the feedback control results, as shown in Figure 3.

[0100] In step S2 (predictive distribution calculation step), the predictive distribution calculation unit 102 uses the result data obtained in S1 to calculate the predictive distribution of a function that shows the relationship between the control parameters to be optimized and the results of feedback control to which those control parameters are applied. The predictive distribution is expressed by the above formulas (1) to (3). Note that in subsequent S2 processes, the previously calculated predictive distribution is updated.

[0101] In S3 (search step), the search unit 103 searches for candidate optimal values ​​for the control parameters based on the predictive distribution calculated in S2. More specifically, the search unit 103 searches for the control parameter values ​​that maximize the acquisition function. This process is represented by the above-mentioned equations (4) and (5). If the reception unit 104 accepted the optimization settings in S1, the search unit 103 performs the search in S3 according to those settings.

[0102] In S4, the search unit 103 outputs the results of the search performed in S3. For example, the search unit 103 may present the detected candidates to the user by displaying a screen as shown in Figure 4.

[0103] In addition, in S4, reference information such as that shown in Figure 4 may also be presented. In this case, the predicted value presentation unit 105 presents a predicted value calculated using the mean and variance of the evaluation value when the candidate detected in S3 is applied. The reason presentation unit 106 then determines the reason for the proposal using the mean and variance mentioned above and presents the determined reason for the proposal. The change pattern presentation unit 107 then identifies the change pattern of the manipulated variable (MV in Figure 2) based on the difference between the comparison target value (typically the latest control parameter shown in the result data obtained in S1) and the candidate detected in S3, and presents information indicating the identified change pattern.

[0104] In S5, the update necessity determination unit 108 determines whether the control parameters applied by the control device 3 have been changed. For example, the update necessity determination unit 108 may confirm whether the control parameters have been changed by communicating with the control device 3, or it may determine that the control parameters have been changed when the user inputs an operation to the information processing device 1 indicating that the control parameters have been changed. If the result in S5 is YES, the process proceeds to S6; if the result in S5 is NO, the process in Figure 6 ends.

[0105] Furthermore, for example, if the information processing device 1 and the control device 3 are located in separate locations and operated by different users, the search unit 103 may notify the terminal device used by the user of the control device 3 of the candidates detected in S3, prompting the user to change the control parameters. In this case, the user of the control device 3 who has seen the notification can then perform an operation to change the control parameters on the control device 3. In this case, the update necessity determination unit 108 may determine that the control parameters have been changed based on the transmission of the above notification.

[0106] In S6, the update necessity determination unit 108 determines whether there are any problems with the operation of the device 4 after the control parameters have been changed. For example, the update necessity determination unit 108 may read the measured value (PV in Figure 2) after the control parameters have been changed from the database 2 and make the determination in S6 based on whether the measured value is within a predetermined normal range.

[0107] If the result in S6 is YES, the process proceeds to S8; if the result in S6 is NO, the process proceeds to S7. In S7, the update necessity determination unit 108 causes the control parameters to be changed. For example, the update necessity determination unit 108 may cause the control parameters to be changed by presenting a message to the user prompting them to change the control parameters. Alternatively, for example, the update necessity determination unit 108 may cause the control parameters to be changed by communicating with the control device 3. In this case, the update necessity determination unit 108 may, for example, instruct the control device 3 to revert the control parameters back to their previous values.

[0108] Furthermore, the process in S6 may be repeated within a predetermined period (e.g., 1 hour) after the control parameters have been changed. This minimizes the impact if the candidate detected in S3 is not valid and causes an abnormality in the operation of device 4.

[0109] The process in S8 is performed after a predetermined period has elapsed, that is, after a change in the control parameters has been confirmed, i.e., after a YES determination in S5. This predetermined period can be set appropriately according to the usage of the device 4, for example, it may be 24 hours. In S8, the update necessity determination unit 108 checks the measured values ​​measured during the predetermined period ( PV ) is retrieved from database 2.

[0110] In S9, the update necessity determination unit 108 uses the measured value acquired in S8 ( PV The result data is generated using the measured values ​​( PV Alternatively, evaluation values ​​may be calculated from the above, and the modified control parameters may be associated with the calculated evaluation values ​​to generate result data.

[0111] In S10, the update necessity determination unit 108 determines whether or not to update the predictive distribution based on the evaluation value included in the result data generated in S9. For example, the update necessity determination unit 108 may determine that an update of the predictive distribution is necessary if the evaluation value meets a predetermined threshold. If NO is determined in S10, the process returns to S8; if YES is determined in S10, the process returns to S1. After this, in S1, the data acquisition unit 101 acquires the result data generated in S9, and in the following S2, the predictive distribution calculation unit 102 updates the predictive distribution using the result data acquired in S1 and the result data generated in S9.

[0112] As described above, the search method executed by the information processing device 1 includes a predictive distribution calculation step (S2) which calculates a predictive distribution of a function that shows the relationship between the control parameters in feedback control and the result of the feedback control to which the control parameters are applied, and a search step (S3) which searches for candidate optimal values ​​for the control parameters based on the predictive distribution calculated in S2. Therefore, it becomes possible to perform stable feedback control without being influenced by human subjectivity.

[0113] [Variation] In the above-described embodiment, an example of searching for candidate optimal values ​​for control parameters in PID control was explained. However, the information processing device 1 can search for candidate optimal values ​​for any control parameter in any feedback control other than PID control.

[0114] Furthermore, the entity executing each process described in the above-described embodiment is arbitrary and not limited to the examples given. In other words, as long as each process described in the above-described embodiment can be executed, the devices constituting the control system 9 can be changed as appropriate. For example, the entity executing each process shown in Figure 6 does not necessarily have to be a single device; these processes can be divided and executed by multiple arbitrary information processing devices (computers).

[0115] [Examples of implementation using software] The function of the information processing device 1 (hereinafter referred to as "the device") is a program that causes the device to function as a computer, and can be realized by a program (search program) that causes each control block of the device (especially each part included in the control unit 10) to function as a computer.

[0116] In this case, the device includes a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., memory) as hardware for executing the program. By executing the program using this control device and storage device, the functions described in each of the embodiments are realized.

[0117] The above program may be recorded on one or more computer-readable recording media, not temporary ones. These recording media may or may not be provided by the above device. In the latter case, the program may be supplied to the above device via any wired or wireless transmission medium.

[0118] Furthermore, some or all of the functions of each of the above control blocks can also be realized by logic circuits. For example, an integrated circuit in which logic circuits functioning as each of the above control blocks are formed is also included in the scope of the present invention. In addition, it is also possible to realize the functions of each of the above control blocks by, for example, a quantum computer.

[0119] The present invention is not limited to the embodiments described above, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention.

[0120] 〔summary〕 An information processing device according to Embodiment 1 of the present invention comprises a predictive distribution calculation unit that calculates a predictive distribution of a function showing the relationship between control parameters in feedback control and the result of feedback control to which the control parameters are applied, and a search unit that searches for candidate optimal values ​​for the control parameters based on the predictive distribution.

[0121] In the information processing device according to aspect 2 of the present invention, the feedback control may be configured to be a feedback control applied in a waste incineration facility, as described in aspect 1 above.

[0122] An information processing device according to embodiment 3 of the present invention may include, in embodiment 1 or 2 above, a data acquisition unit that acquires result data indicating the result of feedback control to which the candidate detected by the search unit is applied, and the predictive distribution calculation unit may be configured to update the predictive distribution using the candidate and the result data.

[0123] The information processing device according to aspect 4 of the present invention may be configured such that, in any of the above aspects 1 to 3, it includes a receiving unit that accepts a selection of whether to prioritize the search for candidates around a known value or within an unknown range of the feedback control results, and the search unit searches for candidates according to the selection accepted by the receiving unit.

[0124] In any of the above embodiments 1 to 4, the information processing device according to embodiment 5 of the present invention may be configured to use an evaluation value as information indicating the result of the feedback control, which is the percentage of measured values ​​that fall within a predetermined tolerance range among the measured values ​​measured during the period in which the feedback control was performed.

[0125] In the information processing device according to embodiment 6 of the present invention, in embodiment 5 described above, the predetermined allowable range may be configured to be a range from the value obtained by subtracting the standard deviation of the measured values ​​measured during the period in which the feedback control was performed from the average value of the measured values ​​measured during that period, to the value obtained by adding the standard deviation to the average value.

[0126] The information processing device according to embodiment 7 of the present invention may be configured to include an update necessity determination unit that, in any of embodiments 1 to 6 above, acquires the result of feedback control to which the candidate detected by the search unit is applied, and repeatedly performs a process at predetermined intervals to determine whether or not to update the predicted distribution based on the acquired result.

[0127] An information processing device according to aspect 8 of the present invention may be configured to include a predicted value presentation unit that presents a predicted value of the result of feedback control to which the candidate detected by the search unit is applied, calculated from the mean and variance of the predicted distribution, in any of the above aspects 1 to 7.

[0128] The information processing device according to aspect 9 of the present invention may be configured to include a reason presentation unit that, in any of the above aspects 1 to 8, presents the reason why the candidate detected by the search unit is valid, based on the mean and variance of the predictive distribution.

[0129] The information processing device according to aspect 10 of the present invention, in any of aspects 1 to 9 above, includes, in any of the above embodiments, a control parameter which includes at least one of a coefficient multiplied by the deviation between a set value and a measured value in PID (Proportional Integral Differential) control, a coefficient multiplied by the integral of the said deviation, and a coefficient multiplied by the derivative of the said deviation, and the feedback control which calculates an operandi for the controlled device from the said deviation and the coefficient, and the device may be configured to include a change pattern presentation unit which presents information showing how the operandi changes when the candidate detected by the search unit based on the updated predicted distribution is applied, based on the difference between the value of the coefficient to be compared and the candidate.

[0130] A control system according to aspect 11 of the present invention includes a control device that performs feedback control on a device to be controlled, and an information processing device that searches for candidate optimal values ​​of the control parameters based on a predictive distribution of a function showing the relationship between the control parameters in the feedback control and the result of the feedback control to which the control parameters are applied, wherein the control device is configured to perform feedback control on the device to be controlled by applying the candidate.

[0131] A search method according to aspect 12 of the present invention is a search method performed by one or more information processing devices, comprising: a predictive distribution calculation step of calculating a predictive distribution of a function that shows the relationship between control parameters in feedback control and the result of feedback control to which the control parameters are applied; and a search step of searching for candidate optimal values ​​of the control parameters based on the predictive distribution.

[0132] A search program according to aspect 13 of the present invention is a search program for causing a computer to function as an information processing device as described in aspect 1 above, and is a search program for causing a computer to function as the predictive distribution calculation unit and the search unit. [Explanation of symbols]

[0133] 1. Information Processing Device 101 Data Acquisition Unit 102 Predictive Distribution Calculation Unit 103 Search Department 104 Reception Department 105 Prediction Value Display Section 106 Reason presentation part 107 Change Mode Presentation Section 108 Update necessity determination unit 3. Control device 9. Control System

Claims

1. A predictive distribution calculation unit calculates a predictive distribution of a function that shows the relationship between control parameters in feedback control and the result of feedback control to which said control parameters are applied. A search unit that searches for candidate optimal values ​​for the control parameters based on the predicted distribution, The search for the aforementioned candidate includes a receiving unit that accepts a selection of whether to prioritize the search in the vicinity of a known value based on the feedback control result, or in the range where the feedback control result is unknown. The search unit is an information processing device that searches for the candidates according to the selection received by the reception unit.

2. The information processing apparatus according to claim 1, wherein the feedback control is a feedback control applied in a waste incineration facility.

3. The system includes a data acquisition unit that acquires result data indicating the result of feedback control applied to the candidate detected by the search unit, The information processing apparatus according to claim 1 or 2, wherein the predictive distribution calculation unit updates the predictive distribution using the candidates and the result data.

4. A predictive distribution calculation unit that calculates a predictive distribution of a function showing the relationship between control parameters in feedback control and the result of feedback control to which the control parameters are applied, The system includes a search unit that searches for candidate optimal values ​​for the control parameters based on the predicted distribution, An information processing device that uses an evaluation value as information indicating the result of the feedback control, which shows the percentage of measured values ​​that fall within a predetermined tolerance range among the measured values ​​measured during the period in which the feedback control was performed.

5. The information processing apparatus according to claim 4, wherein the predetermined tolerance range is a range from the value obtained by subtracting the standard deviation of the measured values ​​measured during the period in which the feedback control was performed from the average value of the measured values ​​measured during that period, to the value obtained by adding the standard deviation to the average value.

6. A predictive distribution calculation unit that calculates a predictive distribution of a function showing the relationship between control parameters in feedback control and the result of feedback control to which the control parameters are applied, A search unit that searches for candidate optimal values ​​for the control parameters based on the predicted distribution, An information processing apparatus comprising: an update necessity determination unit that repeatedly performs a process at predetermined intervals to obtain the result of feedback control to which the candidate detected by the search unit is applied, and to determine whether or not to update the predicted distribution based on the obtained result.

7. A predictive distribution calculation unit that calculates a predictive distribution of a function showing the relationship between control parameters in feedback control and the result of feedback control to which the control parameters are applied, A search unit that searches for candidate optimal values ​​for the control parameters based on the predicted distribution, An information processing apparatus comprising: a prediction value presentation unit that presents a prediction value of the result of feedback control applied to the candidate detected by the search unit, calculated from the mean and variance of the prediction distribution.

8. A predictive distribution calculation unit that calculates a predictive distribution of a function showing the relationship between control parameters in feedback control and the result of feedback control to which the control parameters are applied, A search unit that searches for candidate optimal values ​​for the control parameters based on the predicted distribution, An information processing apparatus comprising: a reason presentation unit that presents the reason why the candidate detected by the search unit is valid, based on the mean and variance of the predictive distribution.

9. A predictive distribution calculation unit that calculates a predictive distribution of a function showing the relationship between control parameters in feedback control and the result of feedback control to which the control parameters are applied, The system includes a search unit that searches for candidate optimal values ​​for the control parameters based on the predicted distribution, The control parameter includes at least one of the following in PID (Proportional Integral Differential) control: a coefficient multiplied by the deviation between the set value and the measured value, a coefficient multiplied by the integral of the said deviation, and a coefficient multiplied by the derivative of the said deviation. The aforementioned feedback control calculates the manipulated variable for the controlled device from the deviation and the coefficient. An information processing device comprising a change mode presentation unit that presents information showing how the manipulated variable changes when the search unit applies the candidate detected based on the updated predicted distribution, based on the difference between the value of the coefficient to be compared and the candidate.

10. A control device that performs feedback control on the controlled device, The system includes an information processing device that searches for candidate optimal values ​​for the control parameters based on a predictive distribution of a function showing the relationship between the control parameters in the feedback control and the result of the feedback control to which the control parameters are applied. The control device applies the candidate and performs feedback control to the controlled device. The information processing device is a control system that, in searching for candidates, accepts a choice as to whether to prioritize the area around a known value for which the result of feedback control is obtained, or the range in which the result of feedback control is unknown, and searches for candidates according to the accepted choice.

11. A control device that performs feedback control on a device to be controlled, The system includes an information processing device that searches for candidate optimal values ​​for the control parameters based on a predictive distribution of a function showing the relationship between the control parameters in the feedback control and the result of the feedback control to which the control parameters are applied. The information processing device is a control system that repeatedly performs, at predetermined intervals, the process of obtaining the result of feedback control by the control device to which the candidate detected by the search is applied, and determining whether or not to update the predicted distribution based on the obtained result.

12. A search method performed by one or more information processing devices, A predictive distribution calculation step involves calculating a predictive distribution of a function that shows the relationship between control parameters in feedback control and the result of feedback control to which said control parameters are applied. A search step of searching for candidate optimal values ​​for the control parameters based on the predictive distribution, The search for the aforementioned candidates includes a receiving step that accepts a choice of whether to prioritize the search in the vicinity of a known value based on the feedback control result, or in the range where the feedback control result is unknown, The search step is a search method that searches for candidates according to the selection received in the reception step.

13. A search program for causing a computer to function as an information processing device according to claim 1, wherein the computer functions as the predictive distribution calculation unit, the search unit, and the reception unit.

Images