Optimal value search controller, optimal value search control method, and computer program

Abstract

To provide an optimal value search controller that avoids degradation in precision of optimal value search control.SOLUTION: An optimal value search controller in accordance with an embodiment includes: an extreme value control unit 100 that is adapted to an arbitrary process of inputting a manipulation quantity U and outputting a measurement value A having a restriction value and an evaluation value Y, searches for an operation point of the manipulation quantity U at which the evaluation value Y measured in a predetermined cycle is inputted and the evaluation value Y becomes an extreme value, and outputs a first adjustment quantity ΔU of the manipulation quantity U; a feedback control unit 300 that inputs the measurement value A, and calculates and outputs a second adjustment quantity ΔU of the manipulation quantity U so that the measurement value A will follow a target value; and a control change mechanism 400 that, when the measurement value A exceeds a threshold, changes inputs from the first adjustment quantity ΔU output from the extreme value control unit 100 to the second adjustment quantity ΔU output from the feedback control unit 300, and supplies the manipulation quantity U after adjustment by the second adjustment quantity ΔU, to the process.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] FIELD Embodiments of the present invention relate to an optimum value search control device, an optimum value search control method, and a computer program. [Background technology]

[0002] In recent years, a technology known as extreme value control has been attracting attention as a plant control method. Extreme value control is a model-free real-time optimal control technology that does not use a complex model of the plant. It calculates the value of the evaluation function (e.g., cost function, performance index, etc.) to be optimized (minimized or maximized) from online sensor information that can directly measure it, and adaptively searches for the evaluation value while changing the manipulated variable so as to maintain it at a (local) optimum value (local minimum value = local minimum value or local maximum value = local maximum value). In other words, the evaluation value representing the plant performance index that is to be maximized (maximized) or minimized (maximized) is measured online.

[0003] The evaluation function is defined as a unified index obtained by multiplying multiple output values ​​of a plant by a predetermined coefficient and then adding them up. In extreme value control, searching for a direction to reduce the evaluation value is equivalent to automatically searching for a direction in which the manipulated variables will reduce all of the outputs (balance at a low value).

[0004] When applying extreme value control to an actual plant, it is important to consider the possibility of situations where constraints set on specific measurement values ​​must be observed. To address such situations, conventional extreme value control employs a penalty function that increases the value of the performance index when the specific measurement value exceeds the constraint. This method prioritizes ensuring that the specific measurement value adheres to the constraint over the optimized behavior achieved by extreme value control. By applying this method, the manipulated variable can be adjusted so that the specific measurement value remains near the constraint value. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Laid-Open No. 2017-33104 Summary of the Invention [Problem to be solved by the invention]

[0006] On the other hand, because extreme value control is a search-type control, the manipulated variable is constantly being driven, resulting in a constant increase and decrease in the manipulated variable. As a result, fluctuations in the manipulated variable cause the measured value to fluctuate around the constraint value, resulting in periods of time in which the constraint is constantly violated. This can lead to a continuous behavior in which the measured value fluctuates around the constraint value. When applied to a water treatment plant, for example, this can lead to the treated water quality deviating from the water quality constraint, causing a decrease in the accuracy of plant control.

[0007] The embodiments of the present invention have been made in consideration of the above circumstances, and provide an optimum value search control device, an optimum value search control method, and a computer program that avoid a decrease in the accuracy of optimum value search control. [Means for solving the problem]

[0008] An optimal value search control device according to an embodiment is applicable to any process that takes an operating variable as an input and outputs a measurement value having a constraint value and an evaluation value, and includes an extreme value control unit that takes the evaluation value measured at a predetermined period as an input, searches for an operating point of the operating variable where the evaluation value becomes an extreme value, and outputs a first adjustment amount of the operating variable; a feedback control unit that takes the measured value as an input, calculates and outputs a second adjustment amount of the operating variable so that the measured value follows a target value; and a control switching mechanism that, when the measured value exceeds a threshold value, switches the input from the first adjustment amount output from the extreme value control unit to the second adjustment amount output from the feedback control unit, and supplies the operating variable adjusted by the second adjustment amount to the process. [Brief explanation of the drawings]

[0009] [Figure 1]FIG. 1 is a block diagram showing an example of the configuration of an optimum value search control device according to an embodiment. [Figure 2] FIG. 2 is a diagram for explaining the principle of extremum control in the optimum value search control device of one embodiment. [Figure 3] FIG. 3 is a diagram for explaining an example of the operation of a control switching mechanism that switches from extreme value control to PID control in an optimum value search control device according to an embodiment. [Figure 4] FIG. 4 is a diagram showing an example of the time variation of the measured value and the manipulated variable when an optimum value search is performed by extremum control. [Figure 5] FIG. 5 is a diagram for explaining an example of the operation of switching control in the optimum value search control device of one embodiment. [Figure 6] FIG. 6 is a diagram for explaining an example of the operation of switching control in the optimum value search control device of one embodiment. [Figure 7] FIG. 7 is a diagram for explaining an example of the operation of switching control in the optimum value search control device of one embodiment. [Figure 8] FIG. 8 is a diagram for explaining an example of the operation of switching control in the optimum value search control device of one embodiment. [Figure 9] FIG. 9 is a diagram showing an example of timing at which the control method for adjusting the manipulated variable is switched from PID control to extreme value control in an optimum value search control device according to an embodiment. [Figure 10] FIG. 10 is a diagram for explaining an example of the operation of a control switching mechanism that switches between extreme value control and PID control in an optimum value search control device according to an embodiment. [Figure 11] FIG. 11 is a block diagram schematically showing another example of the configuration of the optimum value search control device according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, embodiments will be described with reference to the drawings. FIG. 1 is a block diagram showing an example of the configuration of an optimum value search control device according to an embodiment.

[0011] The optimum value search control device of this embodiment is a device equipped with at least one processor and a memory storing a program executed by the processor, and can realize the various functions described below by software or a combination of software and hardware.

[0012] The optimum value search control device of this embodiment outputs an operation amount of the plant 500 so as to optimize an evaluation value based on a measurement value of the plant 500, which is a control target. The optimum value search control device includes an extreme value control unit 100, a gradient estimator 200, a PID control unit (feedback control unit) 300, and a control switching mechanism 400.

[0013] In this embodiment, the plant 500 corresponds to, for example, a sewage treatment plant. When the sludge return rate in the sewage treatment plant is the manipulated variable U, the evaluation value Y corresponds to the total cost calculated by adding the operating costs, such as the cost of pump power, and the water quality cost, which is calculated by converting the water quality, such as the ammonia concentration and phosphorus concentration, of the effluent water into costs. By applying extreme value control, it is possible to search for the sludge return rate (manipulated variable U) that optimizes the total cost, which is the evaluation value Y. The measured value A of the plant 500 corresponds to a process value, such as the phosphorus concentration, that is measured separately in real time in the sewage treatment plant. Note that a sewage treatment plant is one example of the plant 500, and the process (plant) 500 to be controlled is not limited to a sewage treatment plant and can be set according to any process that receives the manipulated variable U as an input and outputs the measured value A and the evaluation value Y.

[0014] The extremum control unit 100 receives an evaluation value Y measured at a predetermined cycle as an input, and outputs an adjustment amount (first adjustment amount) ΔU to be added to the manipulated variable U so as to search for the manipulated variable U at which the evaluation value Y becomes a minimum or maximum. The principle of extremum control will be described later with reference to FIG. 2.

[0015] The gradient estimator 200 has a function of calculating in real time an estimated value of the gradient of the evaluation value Y. The gradient estimator 200 can calculate an estimated value of the gradient of the evaluation value Y in the evaluation function, for example, based on the amount of change ΔY in the evaluation value Y with respect to the amount of operation U per unit time.

[0016] The gradient estimator 200 includes, for example, a storage unit that stores data acquired in real time for a predetermined period, and stores data on the manipulated variable U around the operating point acquired in real time and the evaluation function value acquired in real time for the predetermined period. After accumulating the data, the gradient estimator 200 can calculate an estimate of the gradient of the evaluation function for the manipulated variable U (a function of the evaluation value Y for the manipulated variable U) by linear regression or the like. The period for which data is stored in the gradient estimator 200 (for example, the most recent number of data) is set in advance, and the gradient estimator 200 can update the gradient of the evaluation function by updating the stored data in real time.

[0017] Here, when the gradient estimator 200 estimates the gradient of the performance index function, it is necessary to synchronize (match) the timing of changes in the manipulated variable U and the performance index value. This is because the plant 500 typically has a delay, which means that it takes time for the plant 500 to respond to a change in the manipulated variable U. Even if a data set of the manipulated variable U and the performance index value is acquired as is, the causal relationship between the manipulated variable U and the performance index value may not be properly determined, and the accuracy of the calculated performance index gradient may be reduced. Therefore, to avoid this phenomenon, the gradient estimator 200 may include a delay unit that delays acquisition of the data of the manipulated variable U by a time (first-order lag and dead time) that is aligned with the performance index value. The gradient estimator 200 can calculate the gradient with high accuracy by calculating the gradient using the performance index value and the delayed manipulated variable U.

[0018] The PID control unit 300 is a feedback control unit that receives as input a measurement value signal A that is constrained in the process (plant) 500, and outputs an adjustment amount (second adjustment amount) ΔU to be added to the manipulated variable U so that the measurement value A follows the preset constraint value of the measurement value A. Note that the control in which the measurement value A follows the constraint value is not limited to PID control, and any feedback control that outputs a control amount so that the feedback value follows the target value can be used. In addition to PID control, feedback control that combines the elements of P (proportional), I (integral), and D (differential) can be used depending on the characteristics of the controlled object.

[0019] The control switching mechanism 400 switches between the adjustment amount ΔU of the manipulated variable U output from the extreme value control unit 100 and the adjustment amount ΔU of the manipulated variable U output from the PID control unit 300 as an input value, calculates an adjusted manipulated variable U that reflects the input adjustment amount ΔU, and outputs it to the plant 500. That is, the extreme value control unit 100 and the PID control unit 300 each constantly calculate an adjustment amount ΔU, but the manipulated variable U that is actually input to the plant 500 is obtained by adding one of the adjustment amounts ΔU to the current manipulated variable U by the control switching mechanism 400, and outputting the adjusted manipulated variable U from the control switching mechanism 400.

[0020] The control switching mechanism 400 switches the adjustment amount ΔU that serves as an input value, using information on the signal of the measured value A of the plant 500, the value of the gradient of the evaluation value Y estimated in real time, the current manipulated variable U, and the change trend of the manipulated variable U due to PID control, and calculates and outputs the manipulated variable U that reflects the adjustment amount ΔU. The control switching mechanism 400 includes, for example, a storage unit that stores the manipulated variable U due to PID control for a predetermined period, and can determine the change trend of the manipulated variable U from the value of the manipulated variable U over the predetermined period.

[0021] FIG. 2 is a diagram for explaining the principle of extremum control in the optimum value search control device of one embodiment. The extremum control unit 100 continuously oscillates the input value (operating variable u) of the plant 500 using a dither signal and observes changes (increases or decreases) in the evaluation value (performance function value) of the plant 500. The dither signal used in extremum control is often given as a sine wave. Based on changes in the evaluation value y in response to changes in the operating variable u, the extremum control unit 100 changes the operating variable in a direction that brings the evaluation value closer to the optimal value. The extremum control unit 100 searches for the optimal value of the performance function y by repeating the above-described changes in the operating variable u.

[0022] The extremum control unit 100 includes a high-pass filter (HPF) 101, dither signal generators 102 and 106, a low-pass filter (LPF) 104, an integrator 105, multiplication units 103 and 108, and a coefficient multiplication unit 107.

[0023] The dither signal generator 106 has a function of providing a perturbation signal to be added to the manipulated variable u, and can provide fluctuations to the evaluation function value y. The low-pass filter 104 has a function of extracting low-frequency components of the input signal, and extracts the low-frequency components of the signal obtained by multiplying the evaluation function value y by a dither signal. From the output value of the low-pass filter 104, the extreme value control unit 100 can determine whether the evaluation function value y has increased or decreased.

[0024] The high-pass filter 101 extracts high-frequency components from the evaluation function value y fed back from the plant 500, thereby removing a constant bias corresponding to the optimum value.

[0025] The integrator 105 outputs a value obtained by integrating the low-frequency components extracted by the low-pass filter 104. The output value of the integrator 105 is a total value obtained by adding together the increase and decrease amounts of the evaluation function value y over a predetermined time, and the integrator 105 functions as an estimator that estimates the direction (positive or negative) of the manipulated variable u that should be changed in order to bring the evaluation function value y closer to the optimal value.

[0026] The output value of the integrator 105 is multiplied by the product of the output value of the dither signal generator 106 and the coefficient a of the coefficient multiplier 107, and is supplied to the plant 500 as the manipulated variable u. Based on the above-described principle of extreme value control, the extreme value control unit 100 calculates an adjustment amount ΔU of the manipulated variable of the plant 500 while searching for an optimal value of the evaluation value Y.

[0027] Next, the operation of the control switching mechanism 400 will be described. FIG. 3 is a diagram for explaining an example of the operation of a control switching mechanism that switches from extreme value control to PID control in an optimum value search control device according to an embodiment.

[0028] First, a mechanism for switching control will be described for the case where the control switching mechanism 400 switches the input value from the adjustment amount ΔU by the extreme value control unit 100 to the adjustment amount ΔU by PID control.

[0029] When extreme value control is applied with the sludge return rate in a sewage treatment plant as the manipulated variable U and the total cost as the evaluation value Y, it is possible to search for the sludge return rate (manipulated variable U) that optimizes the total cost, which is the evaluation value Y. As described above, in the process of changing the sludge return rate (manipulated variable U) of the sewage treatment plant using extreme value control, the phosphorus concentration (measured value A) of the treated water, which is measured separately in real time at the sewage treatment plant, changes. Note that here, an example of the change over time between the manipulated variable U and the measured value A will be described in which the measured value A tends to rise (increase) in accordance with fluctuations in the manipulated variable U when the manipulated variable U is calculated using extreme value control.

[0030] For example, if a target value (constraint) is set for the phosphorus concentration, the phosphorus concentration may deviate from the target value during extreme value control. To prevent this situation, the optimum value search control device of this embodiment uses the control switching mechanism 400 to switch the input value from the adjustment amount ΔU by extreme value control to the adjustment amount ΔU by PID control when the measured value A exceeds a threshold value.

[0031] Here, an example will be described in which the measured value A deteriorates (increases) and exceeds the constraint in the process of searching for the optimum value of the manipulated variable U by extremum control. FIG. 4 is a diagram showing an example of the time variation of the measured value and the manipulated variable when an optimum value search is performed by extremum control.

[0032] This shows an example of the time variation of the manipulated variable U and the measured value A when the manipulated variable U is calculated using extreme value control and the measured value A rises in accordance with fluctuations in the manipulated variable U. In this case, the manipulated variable U will be adjusted so that the measured value A stays near the constraint, and the extreme value control behavior will cause the manipulated variable U to fluctuate, potentially resulting in an event where the measured value A oscillates unnecessarily near the constraint.

[0033] Extreme value control is a control that searches for the optimal value of the manipulated variable, but when there is a constraint on the measured value A, the value of the evaluation function increases due to the influence of the penalty function of the measured value A, and at the stage when the measured value A exceeds the constraint, the behavior changes to one that prioritizes preventing the measured value A from deviating from the constraint over searching for the optimal value of the manipulated variable. As a result, as shown in Figure 4, the behavior of the measured value A will remain near the constraint value. However, because extreme value control is a control that searches for the optimal value while actively changing the manipulated variable, this influence will cause the measured value A to continue to fluctuate near the constraint value. When applied to a water treatment plant, for example, this can lead to deviations from water quality constraints in the treated water quality.

[0034] Furthermore, since extreme value control is a very slow control, the period during which the manipulated variable U fluctuates is long, and therefore the period during which the constraints are violated due to unnecessary vibrations in the measured value A caused by fluctuations in the manipulated variable U may become so long that it is unacceptable in practical operation.

[0035] In contrast, according to the optimum value search control device of this embodiment, a control switching threshold is set at a value lower than the constraint of the measured value A, and when the measured value A exceeds the threshold, the control switching mechanism 400 switches to PID control using the adjustment amount ΔU as an input. This eliminates active fluctuations in the manipulated variable due to extreme value control, as shown in FIG. 3, and the measured value A moves to follow the constraint value, thereby preventing unnecessary deviations from the constraint value. Furthermore, by setting the threshold for the measured value A to a value lower than the constraint value, it is possible to prevent the measured value A from deviating from the constraint value immediately before switching from extreme value control to PID control.

[0036] In other words, when a constraint is placed on the measured value A and that constraint must be prioritized in actual operation, the control switching mechanism 400 switches the input value from the adjustment amount ΔU based on extreme value control to the adjustment amount ΔU based on PID control when the measured value A approaches the constraint value.

[0037] After the control switching mechanism 400 switches from extreme value control to PID control, if the measured value A no longer needs to comply with the constraints in the optimal value search control (if the evaluation value Y can converge to the optimal value without making the measured value A follow the constraint value), the control switching mechanism 400 switches the input value from the adjustment amount ΔU by PID control to the adjustment amount ΔU by extreme value control.

[0038] The information used by the control switching mechanism 400 to switch the input value from the adjustment amount ΔU by PID control to the adjustment amount ΔU by extreme value control includes change history information on the manipulated variable U and gradient information on the evaluation value Y, which is the input to the extreme value control.

[0039] Extremum control is a control that searches for the manipulated variable U that minimizes the value of the input evaluation value Y, and the search is carried out while estimating the gradient of the evaluation value Y during the search process. Extremum control is a mechanism that drives the system to correct the manipulated variable U when the gradient of the estimated evaluation value Y has a value (is not zero). Since the direction in which the manipulated variable is corrected is determined by the sign of the evaluation function, by monitoring the gradient information of the evaluation function, it is possible to determine in which direction the manipulated variable should be changed next when extremum control is driven, or to make predictions about the current state.

[0040] The above operation of the control switching mechanism 400 is an example, and the target process is not limited to a sewage treatment plant, and the manipulated variable U, the measured value A, etc. can be set according to the target process.

[0041] The specific control switching operation by the control switching mechanism 400 will be described below. 5 to 8 are diagrams for explaining an example of the operation of switching control in the optimum value search control device of one embodiment.

[0042] Here, we will explain the control switching operation by the control switching mechanism 400 when the evaluation function for the manipulated variable U is expressed as a downward convex function, the slope of the measured value A with respect to the manipulated variable U is negative, and the minimum value of the evaluation function is searched for by extreme value control, and the operating point that fluctuates the manipulated variable U is changed to the negative side.

[0043] 5 to 8, the extreme value control causes the manipulated variable U to change from the positive side to the negative side of the extreme value of the evaluation function value. At this time, the measured value A tends to increase, and when the measured value A exceeds the switching threshold of the control switching mechanism 400 (FIG. 6), the input to the control switching mechanism 400 switches from the adjustment variable ΔU by the extreme value control to the adjustment variable ΔU by the PID control.

[0044] After that, as the operating point of the manipulated variable U moves further to the negative side, the operating point of the manipulated variable U where the measured value A becomes the constraint value of the measured value A stabilizes (Fig. 7). After that, as long as the manipulated variable U is basically adjusted using PID control, the manipulated variable U will be adjusted so that the measured value A remains equal to the constraint value.

[0045] Once control switches from extreme value control to PID control, it is difficult to determine whether switching from PID control to extreme value control is possible based solely on the signal value of measurement value A. Therefore, in the optimum value search control device of this embodiment, the control switching mechanism 400 determines the timing to switch from PID control to extreme value control using the change history of the manipulated variable U and gradient information calculated in real time.

[0046] For example, as shown in Figure 7, if the trend of measured value A relative to the manipulated variable U changes (the line for measured value A shifts downward), if the manipulated variable U is driven by PID control, the operating point of the manipulated variable U will be adjusted downward (toward the negative side) so that the measured value A is equal to the constraint value. In the case of a sewage treatment plant, this corresponds to an improvement in the phosphorus concentration of the effluent water quality due to a reduction in inflow load. In the example of Figure 7, even if extreme value control were applied, the manipulated variable U would be changed in the direction that decreases the evaluation function value, resulting in a similar decrease in the manipulated variable U. Because the direction of change in the manipulated variable U to minimize the evaluation function value is consistent with the direction of change in the manipulated variable U by PID control, in such cases, there is no need to switch from PID control to extreme value control. The estimated gradient of the evaluation function near the operating point of the manipulated variable U is positive. This indicates that when PID control tends to decrease the manipulated variable, it is advisable to continue using PID control when the estimated gradient of the evaluation function near the operating point is positive.

[0047] Next, Fig. 8 shows an example of a graph in which the trend of the measured value A changes further from the state shown in Fig. 7 (the straight line shifts downward). If PID control is applied to search for the optimal value of the manipulated variable U in this state, the manipulated variable U will be adjusted to further decrease so that the measured value A becomes equal to the constraint value. In the process of further decreasing the manipulated variable U, the evaluation function value will increase past the minimum value (the value corresponding to the optimal value of the manipulated variable U). This state satisfies the conditions under which it is theoretically possible to search for the manipulated variable U at which the measured value A satisfies the constraint condition, using extremum control.

[0048] From the above, the timing at which the input of the control switching mechanism 400 is switched from the adjustment amount ΔU based on PID control to the adjustment amount ΔU based on extreme value control is the timing at which the gradient of the evaluation value calculated in real time changes from a positive value to a negative value. In this way, when the gradient of the evaluation value is negative while searching for the optimal value of the manipulated variable U by extreme value control, the manipulated variable U acts to increase, so it is possible to search for the manipulated variable U at which the evaluation function value is minimized without deviating from the constraints of the measured value A.

[0049] Furthermore, if the trend of the measured value A relative to the manipulated variable changes continuously, the manipulated variable U is continuously adjusted by PID control, and the gradient of the evaluation value may also change continuously from negative to positive. In this case, the timing for switching the input of the control switching mechanism 400 from the adjustment amount ΔU by PID control to the adjustment amount ΔU by extreme value control is not when the gradient of the evaluation value becomes even slightly positive, but when the gradient of the evaluation value reaches a predetermined positive value. This is because the point at which the gradient of the evaluation value switches from a negative value to a positive value (when the gradient of the evaluation value becomes 0) is the timing when the manipulated variable U is being driven to achieve a nearly optimal evaluation value. However, if the control method for adjusting the manipulated variable U is switched from PID control to extreme value control in that situation, the manipulated variable U may begin to fluctuate, potentially causing the measured value A to deviate from the constraint value. Therefore, when the gradient of the evaluation value takes a predetermined positive value, by changing the control method from PID control to extreme value control, it is possible to keep the measured value A below the constraint value in the process of searching for the manipulated variable U that minimizes the value of the evaluation function.

[0050] The above example showed behavior when the manipulated variable under PID control tended to decrease, but if extreme value control was switched to PID control in a situation where the operating point was to the left of the optimal value, the setting is made so that the opposite trend holds true in PID control as the manipulated variable tends to increase. This refers to a state where the measured value tends to deteriorate as the manipulated variable increases, and as the measured value A deteriorates, extreme value control is switched to PID control when it exceeds a certain threshold, and then PID control adjusts the manipulated variable so that the measured value = constraint value.

[0051] FIG. 9 is a diagram showing an example of timing at which the control method for adjusting the manipulated variable is switched from PID control to extreme value control in an optimum value search control device according to an embodiment. The control switching mechanism 400 uses the change tendency of the operating point of the manipulated variable U and the sign of the gradient of the evaluation function estimated around the operating point as criteria for switching the control method for adjusting the manipulated variable U from PID control to extreme value control.

[0052] If the gradient of the evaluation function is positive and the change trend of the manipulated variable U by PID control is increasing, the control switching mechanism 400 determines that there is no possibility that the measured value A will deviate from the constraint even if the adjustment variable ΔU is calculated by extreme value control, and switches the input value from the adjustment variable ΔU by PID control to the adjustment variable ΔU by extreme value control.

[0053] If the gradient of the evaluation function is positive and the change trend of the manipulated variable U by PID control is decreasing, the control switching mechanism 400 determines that there is a possibility that the measured value A will deviate from the constraint even if the adjustment amount ΔU is calculated by extreme value control, and adjusts the manipulated variable U using the adjustment amount ΔU by PID control without switching the input value.

[0054] If the gradient of the evaluation function is negative and the change trend of the manipulated variable U by PID control is increasing, the control switching mechanism 400 determines that there is a possibility that the measured value A will deviate from the constraint even if the adjustment amount ΔU is calculated by extreme value control, and adjusts the manipulated variable U using the adjustment amount ΔU by PID control without switching the input value.

[0055] If the gradient of the evaluation function is negative and the change trend of the manipulated variable U by PID control is decreasing, the control switching mechanism 400 determines that there is no possibility that the measured value A will deviate from the constraint even if the adjustment variable ΔU is calculated by extreme value control, and switches the input value from the adjustment variable ΔU by PID control to the adjustment variable ΔU by extreme value control.

[0056] FIG. 10 is a diagram for explaining an example of the operation of a control switching mechanism that switches between extreme value control and PID control in an optimum value search control device according to an embodiment. Here, we will explain an example of the time variation of the manipulated variable U and the measured value A when the manipulated variable U is calculated by extreme value control and the measured value A tends to rise (increase) in accordance with the fluctuation of the manipulated variable U. For example, a constraint is set on the measured value A.

[0057] In the optimum value search control device of this embodiment, when the measurement value A exceeds a threshold value (constraint threshold value), the control switching mechanism 400 switches the input value from the adjustment amount ΔU by extreme value control to the adjustment amount ΔU by PID control.

[0058] When the manipulated variable U is adjusted by the adjustment amount ΔU by PID control, the measured value A changes to follow the constraint value and the measured value A will no longer exceed the constraint value. At this time, if the gradient of the evaluation function is positive and the manipulated variable U is not on an increasing trend, the control switching mechanism 400 continues to adjust the manipulated variable U using the adjustment amount Δ by PID control without switching the input value.

[0059] Thereafter, when the gradient of the evaluation function becomes negative, the control switching mechanism 400 switches the input value from the adjustment amount ΔU by PID control to the adjustment amount ΔU by extreme value control. As shown in Fig. 10, the control switching mechanism 400 reads the sign of the gradient of the evaluation function in real time, so that the control method for adjusting the manipulated variable U can be switched from PID control to extreme value control at an appropriate timing, and the optimal manipulated variable U can be searched for by extreme value control.

[0060] If the fluctuation of the measurement value A is large, the control switching mechanism 400 may calculate a moving average or the like of the measurement value A in order to grasp the average behavior direction of the measurement value A. Similarly, if the fluctuation of the gradient of the evaluation function at the operating point is also large, the control switching mechanism 400 may calculate a moving average or the like of the gradient of the evaluation function and determine the change trend from the moving average or the like.

[0061] As described above, according to this embodiment, when extreme value control is applied to an actual plant, in cases where operation is required to comply with constraints on specific measurement values, it is possible to avoid a period of deviation from the constraints becoming so long that it is operationally unacceptable, and it is possible to provide an optimal value search control device, an optimal value search control method, and a computer program that avoid a decrease in the accuracy of the optimal value search control.

[0062] FIG. 11 is a block diagram schematically showing another example of the configuration of the optimum value search control device according to an embodiment. The optimum value search control device shown in FIG. 11 is a modified example of the optimum value search control device shown in FIG. 1, and differs from the optimum value search control device shown in FIG. 1 in that it includes a linear regression type extreme value control unit 110 and omits the gradient estimator 200.

[0063] For example, in the configuration of the extreme value control unit 100 shown in FIG. 2, the output of the low-pass filter 104 corresponds to the gradient of the evaluation function estimated by the extreme value control unit 100, and is a signal that can be used as the gradient of the evaluation function value used to determine the timing of switching the control method in the control switching mechanism 400.

[0064] The output value of the low-pass filter 104 has a certain degree of accuracy as an estimated value of the gradient of the evaluation function when operating under extreme value control. However, even when the manipulated variable U is being adjusted by PID control, the calculation itself within the extreme value control unit 100 continues, and the output value of the low-pass filter 104 cannot be used as the gradient value of the evaluation function during the period when the manipulated variable U is being adjusted by PID control. This is because the estimation of the gradient of the evaluation function in the extreme value control unit 100 requires that the evaluation function fluctuates in accordance with the periodic fluctuation of the manipulated variable U.

[0065] 11, in order to improve the convergence of the extremum control, an extremum control unit 110 having a mechanism for estimating a gradient using linear regression or the like is provided, and the gradient value (or a value equivalent to the gradient value) of the evaluation function calculated inside the extremum control unit 110 can be used to determine the timing for switching the control method in the control switching mechanism 400. As a result, in the optimal value search control device of this modification, the control method can be switched using an internal signal indicating the gradient value of the extremum control unit 110, and there is no need to provide a separate gradient estimator for the evaluation function.

[0066] In this modified example, the signal of the manipulated variable U input to the extreme value control unit 110 is the output signal of the control switching mechanism 400 rather than the output signal of the extreme value control unit 110, and processing calculations by the extreme value control unit 110 are always continued even during the period when the control method for adjusting the manipulated variable U is switched to PID control.

[0067] According to the above-described modified example, the same effect as that of the above-described present embodiment can be obtained, and when extreme value control is applied to an actual plant, in cases where operation is required to comply with constraints on specific measurement values, it is possible to avoid a period of deviation from the constraints becoming so long that it is operationally unacceptable, and it is possible to provide an optimal value search control device, an optimal value search control method, and a computer program that avoid a decrease in the accuracy of the optimal value search control.

[0068] The various functions of the optimum value search control device of the above-described embodiment and modified example may be realized by hardware or software. Furthermore, when the various functions of the optimum value search control device are realized by software, the optimum value search control device may be configured to store setting information such as the dither signal applied to each manipulated variable in extreme value control and the normalization processing function in a storage device such as a magnetic hard disk drive or a semiconductor storage device, and to acquire the setting information from the storage device. In this case, the optimum value search control device may be configured to include an input unit that accepts changes to the dither signal and the normalization processing, and a setting update unit that updates the setting information for the dither signal and the normalization processing in response to the input. Furthermore, in this case, the optimum value search control device may be configured to include a display unit that displays the contents of the setting information.

[0069] The program according to this embodiment may be transferred in a state where it is stored in an electronic device, or in a state where it is not stored in an electronic device. In the latter case, the program may be transferred via a network, or in a state where it is stored in a storage medium. The storage medium is a non-transitory tangible medium. The storage medium is a computer-readable medium. The storage medium may be in any form, such as a CD-ROM or a memory card, as long as it is capable of storing the program and is computer-readable.

[0070] Although several embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be embodied in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included within the scope and spirit of the invention, and are also included in the scope of the invention and its equivalents as defined in the claims. [Explanation of symbols]

[0071] 100...Extreme value control unit, 101...High-pass filter, 102...Dither signal generator, 103...Multiplication unit, 104...Low-pass filter, 105...Integrator, 106...Dither signal generator, 107...Coefficient multiplication unit, 108...Multiplication unit, 110...Linear regression type extreme value control unit, 200...Slope estimator, 300...PID control unit (feedback control unit), 400...Control switching mechanism, 500...Plant (process)

Claims

1. It is applicable to any process that takes manipulated variables as inputs and outputs measured values ​​with constraints and evaluation values, an extreme value control unit that receives the evaluation value measured at a predetermined cycle as an input, searches for an operating point of the manipulated variable at which the evaluation value becomes an extreme value, and outputs a first adjustment amount of the manipulated variable; a feedback control unit that receives the measurement value as an input, calculates a second adjustment amount for the manipulated variable so that the measurement value follows a target value, and outputs the second adjustment amount; and a control switching mechanism that, when the measurement value exceeds a threshold value, switches input from the first adjustment amount output from the extreme value control unit to the second adjustment amount output from the feedback control unit, and supplies the manipulated variable after adjustment by the second adjustment amount to the process.

2. 2. The optimum value search control device according to claim 1, wherein the control switching mechanism determines whether or not there is a possibility that the measurement value of the process will deviate from a constraint based on the measurement value of the process, a history of the manipulated variable input to the process, and gradient information of the evaluation value with respect to the manipulated variable, and when it is determined that the measurement value of the process will not deviate from the constraint, switches the input from the second adjustment variable output from the feedback control unit to the first adjustment variable output from the extreme value control unit, and supplies the manipulated variable adjusted by the first adjustment variable to the process.

3. 3. The optimum value search control device according to claim 2, wherein the gradient information of the evaluation value with respect to the manipulated variable uses an internal signal calculated in the extreme value control unit.

4. 3. The optimum value search control device according to claim 2, further comprising a gradient estimator that calculates, as the gradient information, an estimated value of the gradient of a function of the evaluation value with respect to the manipulated variable, using data of the manipulated variable and the evaluation value for a predetermined period around the operating point.

5. It is applicable to any process that takes manipulated variables as inputs and outputs measured values ​​with constraints and evaluation values, when the measured value exceeds a threshold, an adjustment amount for adjusting the manipulated variable is switched from a first adjustment amount for the manipulated variable calculated by searching for an operating point of the manipulated variable at which the evaluation value becomes an extreme value to a second adjustment amount for the manipulated variable calculated so that the measured value follows a target value; and supplying the manipulated variable after adjustment by the second adjustment variable to the process.

6. A computer program that causes a computer to function as the optimum value search control device according to claim 1.

Images