Control performance evaluation method, control performance evaluation program, control performance evaluation device, and model performance evaluation method

Abstract

This invention provides a control performance evaluation method that can calculate performance indicators related to predictive control having a process model that shows the dynamic characteristics of the controlled object. [Solution] The control performance evaluation method involves acquiring data measured or calculated for predictive control using a process model that shows the dynamic characteristics of the controlled object, which includes a predicted value of the controlled amount of the controlled object predicted using the process model that shows the dynamic characteristics of the controlled object, and a trajectory value for tracking the set value. A computer then performs a process to calculate a control performance index based on the predicted value and the trajectory value.

Description

[Technical Field]

[0001] The present invention relates to a control performance evaluation method, a control performance evaluation program, a control performance evaluation device, and a model performance evaluation method. [Background technology]

[0002] In petroleum and chemical plants, numerous controllers, including PID control systems, are in operation, but it is not easy for a person to visually grasp the control status of all these controllers. A control performance evaluation method that uses minimum variance control as a benchmark is known as a method for evaluating the control performance of controllers from plant operation data, and is used for monitoring controllers operating in plants and factories (see Non-Patent Document 1). The aforementioned control performance evaluation method mainly targets controllers with one input and one output, but control performance evaluation methods targeting multivariable model predictive control and methods for maintaining model predictive control have also been proposed (for example, Patent Document 1). In addition to the aforementioned model predictive control, predictive control that has a process model showing the dynamic characteristics of the controlled object includes PFC (Predictive Functional Control) (Non-Patent Document 2) and internal model control. [Prior art documents] [Patent Documents]

[0003] [Patent Document 1] Special Publication No. 2010-522942 [Non-patent literature]

[0004] [Non-Patent Document 1] Manabu Kano, Yoshiyuki Yamashita, "Evaluation and Monitoring of Control Performance of Process Control Systems," Measurement and Control, 2002. [Non-Patent Document 2] Jacques Richalet, Donal O'Donovan, “Predictive Functional Control”, Springer, 2009 [Overview of the project] [Problems that the invention aims to solve]

[0005] One aspect of this is to provide a control performance evaluation method that can calculate performance indicators related to predictive control, which has a process model that shows the dynamic characteristics of the controlled object. [Means for solving the problem]

[0006] The control performance evaluation method involves acquiring data measured or calculated for predictive control using a process model that shows the dynamic characteristics of the controlled object, which includes a predicted value of the controlled quantity of the controlled object predicted using the process model that shows the dynamic characteristics of the controlled object, and a trajectory value for tracking the set value. A computer then performs a process to calculate a control performance index based on the predicted value and the trajectory value. A control performance evaluation method, wherein a computer performs a process to calculate the control performance index by determining the ratio of the variance of the difference between the predicted value and the trajectory value to the minimum variance of the difference. . [Effects of the Invention]

[0007] In one respect, it is possible to calculate performance metrics related to predictive control that have a process model that shows the dynamic characteristics of the controlled object. [Brief explanation of the drawing]

[0008] [Figure 1] This is an explanatory diagram showing an example configuration of a control performance evaluation system. [Figure 2] This is a block diagram showing an example configuration of a control performance evaluation device. [Figure 3] This is a block diagram showing the structure of the PFC. [Figure 4] This is an explanatory diagram regarding control performance indicators. [Figure 5] This is an explanatory diagram regarding model performance metrics. [Figure 6] This figure shows an example of control performance evaluation results. [Figure 7] This figure shows an example of a trend graph. [Figure 8] This flowchart shows an example of the processing procedure performed by a control performance evaluation device. [Modes for carrying out the invention]

[0009] The present invention will be described in detail below with reference to the drawings illustrating its embodiments. (Embodiment) Figure 1 is an explanatory diagram showing an example configuration of a control performance evaluation system. In this embodiment, a control performance evaluation system is described that evaluates the controllability of a PFC and the validity of a process model by calculating a PFC control performance index and a model performance index of a process model that simulates the dynamic characteristics of a controlled object from operating data of a controlled object controlled by the PFC method. The control performance evaluation system includes a control performance evaluation device 1, a plant information management device 2, a terminal 3, an OPC (OLE for Process Control) server 4, and a DCS (Distributed Control System) 5. The control performance evaluation device 1, the plant information management device 2, the terminal 3, and the OPC server 4 are connected via network N1, and the DCS 5 and the OPC server 4 are connected via network N2.

[0010] Predictive control using a process model that shows the dynamic characteristics of the controlled object is a control method that predicts the future using the response from the controlled object (controlled variable) and a process model that shows the dynamic characteristics of the controlled object, and determines the manipulated variable so that the predicted value approaches the target value (setpoint). In model predictive control and PFC, a curve called a reference trajectory is defined that brings the controlled variable closer to the setpoint, and the manipulated variable is controlled so that the difference between the predicted value of the controlled variable and the reference trajectory is minimized (see Figure 4, etc.). In this embodiment, we will describe a form of PFC that assumes that the controlled object process has a first-order lag and dead time.

[0011] In this embodiment, we will describe a configuration in which PFC is employed as a predictive control method having a process model that shows the dynamic characteristics of the controlled object. However, predictive control having a process model that shows the dynamic characteristics of the controlled object is not limited to PFC; any method that uses a process model that shows the dynamic characteristics of the controlled object to perform control is acceptable.

[0012] The control performance evaluation device 1 is a device capable of various information processing and information transmission / reception, such as a server computer or personal computer. The control performance evaluation device 1 calculates control performance indicators for evaluating the controllability of the PFC and model performance indicators for evaluating the validity of the process model from the operating data of the controlled object measured or calculated for control by the PFC. The controlled object is not particularly limited, but for example, production equipment in a chemical plant is envisioned. The control performance evaluation device 1 evaluates the performance based on each calculated indicator and outputs model parameters to be set in the process model as recommended values ​​according to the evaluation results.

[0013] The plant information management device 2 is a database server that stores time-series data such as operating data of the controlled object controlled using the PFC method, and the control performance evaluation device 1 acquires the operating data of the controlled object from the plant information management device 2, calculates each indicator, and outputs the evaluation results to terminal 3.

[0014] In this embodiment, the control performance evaluation device 1 is assumed to be an online or offline system installed on a local network that performs control performance evaluation. However, the system may also be configured as an online monitoring system virtually located on the cloud.

[0015] DCS5 is a device that controls the controlled object. DCS5 calculates the manipulated variables according to the PFC algorithm and controls the controlled object. Plant information management device 2 acquires operating data of the controlled object from DCS5 via OPC server 4 and transmits it to control performance evaluation device 1. The evaluation results from control performance evaluation device 1 are output to terminal 3.

[0016] Figure 2 is a block diagram showing an example configuration of the control performance evaluation device 1. The control performance evaluation device 1 comprises a control unit 11, a main memory unit 12, a communication unit 13, and an auxiliary memory unit 14. The control unit 11 has one or more arithmetic processing units such as a CPU (Central Processing Unit), MPU (Micro-Processing Unit), and GPU (Graphics Processing Unit), and performs various information processing by reading and executing the program P stored in the auxiliary storage unit 14. The main memory unit 12 is a temporary storage area such as SRAM (Static Random Access Memory) or DRAM (Dynamic Random Access Memory), and temporarily stores the data necessary for the control unit 11 to perform arithmetic processing. The communication unit 13 is a communication module for performing communication-related processing, and transmits and receives information with the outside. The auxiliary storage unit 14 is a non-volatile storage area such as a hard disk or large-capacity memory, and stores the program P (program product) and other data necessary for the control unit 11 to perform processing.

[0017] The control performance evaluation device 1 may be a multicomputer consisting of multiple computers, or it may be a virtual machine virtually constructed by software.

[0018] Furthermore, in this embodiment, the control performance evaluation device 1 is not limited to the above configuration and may include, for example, an input unit for receiving operation inputs, a display unit for displaying images, etc. Also, the control performance evaluation device 1 may include a reading unit for reading portable storage media 1a such as CD (Compact Disk)-ROM or DVD (Digital Versatile Disc)-ROM, and may read and execute a program P from the portable storage media 1a.

[0019] Figure 3 is a block diagram showing the structure of the PFC. Figure 3 illustrates the structure of the PFC assumed in this embodiment. The outline of this embodiment will be described below.

[0020] PFC has the structure of internal model control that accounts for dead time compensation. As described above, the present embodiment assumes a PFC that assumes the dynamic characteristics of a controlled object to be represented by a first-order lag and a dead time. Note that the process model is not limited to those including a first-order lag and a dead time, and may include, for example, process models including integration and those including a second-order lag.

[0021] First, the basic operation of PFC will be described. In PFC, a predicted value of the controlled variable PV of the controlled object subjected to dead time compensation (PV in FIG. 3 pred ) and a setpoint SV are used to calculate a reference trajectory y r , and the manipulated variable MV to the controlled object is determined such that the predicted value PV pred matches the reference trajectory y r to the controlled object.

[0022] In PFC, the manipulated variable MV determined above is input to the controlled object to control the controlled object, and the controlled variable PV is measured. Further, by inputting the manipulated variable MV output to the controlled object into a process model that indicates the dynamic characteristics of the controlled object and takes dead time into account (hereinafter referred to as "first process model"), an output value y of the first process model h is obtained. At the same time, by inputting the manipulated variable MV into a process model that does not include dead time (hereinafter referred to as "second process model"), an output value y of the process model before dead time m is obtained.

[0023] Then, using the output value y of the first process model h and the output value y of the process model before dead time m , a predicted value PV of the controlled variable PV compensated for dead time pred is calculated. In the next control cycle, the manipulated variable MV is determined such that the calculated predicted value PV pred matches the reference trajectory y r .

[0024] In PFC, the above processing is repeated to control the controlled object.

[0025] The specific processing details will be explained using mathematical formulas. First, when evaluating control performance, model performance, or calculating recommended parameters, if the operating data of the controlled object does not contain a sufficient amount of data, the accuracy may deteriorate. For example, the controlled variable PV or the predicted value PV. pred Furthermore, if the manipulated variable MV is stable data with little change, small fluctuations that are not practically problematic may be captured, leading to a misjudgment of poor control performance. Additionally, because the identification is performed using data with limited information, there is a risk of calculating incorrect recommended parameters.

[0026] Therefore, the control performance evaluation device 1 predicts the value PV pred The following series of operations are performed only if the variance of is greater than or equal to the third threshold γ3, and the variance of the manipulated variable MV is greater than or equal to the fourth threshold γ4. Note that the process gain K of the process model m Alternatively, threshold γ3 can be obtained from threshold γ4 using the following equation (1). Alternatively, threshold γ4 can be obtained from threshold γ3 using the relationship in equation (1). Furthermore, the predicted value PV pred Alternatively, the evaluation can be performed using the controlled variable PV.

[0027]

number

[0028] Next, we will explain the PFC algorithm. First, the discrete-time model with a first-order lag is expressed by the following equation (2).

[0029]

number

[0030] Here, y m This is the output value of the second process model, which does not include dead time, and K m This is the process gain, and T m is the time constant, T srepresents the sampling time. Using this model, the predicted value of the control variable at time k, h time (h steps) ahead, can be calculated as shown in equation (3) below.

[0031]

number

[0032] y m (k+h|k) represents the predicted value of the controlled variable at time k, h time ahead. Δy m (k+h|k)=y m (k+h|k)-y m If we let (k), then from equation (3), Δy m (k+h|k) is given by the following equation (4).

[0033]

number

[0034] Next, let's consider dead time compensation. If there is dead time d in the process, the process model can be expressed as follows.

[0035]

number

[0036] Here, y h This is the output value of the first process model, including the dead time d. In PFC, dead time compensation is performed using these process models. The controlled variable to be controlled is PV, and the predicted value of the controlled variable PV after dead time compensation is PV. pred Therefore, the predicted value PV pred This can be expressed by the following formula (6).

[0037]

number

[0038] As shown in equation (6), the predicted value PV predThis is the output value y from the first process model, derived from the controlled variable PV. h Subtracting this, and further, the output value y from the second process model m It is obtained by adding y. h is y m Since we can simply shift the timeout, equation (6) can be treated as equation (7) below.

[0039]

number

[0040] Next, the reference orbit y r Let's consider this. In PFC, the predicted value of the controlled variable PV is PV. pred A reference trajectory y that gradually asymptotically approaches the set value SV. r Given a given, the manipulated variable MV is calculated accordingly. The reference trajectory y at time k is at time h. r This can be expressed by the following formula (8).

[0041]

number

[0042] Here, TRBF is a control parameter, and generally, the time constant T is used as the initial value. m A static determination time three times the given time is given by Δy r (k+h|k)=y r (k+h|k)-PV pred Let (k) be Δy r (k+h|k) can be expressed by the following equation (9).

[0043]

number

[0044] Next, let's consider the formula for the manipulated variable MV. In PFC, Δy r (k+h|k) and Δy mThe manipulated variable MV is calculated such that (k+h|k) is equal. From equations (4) and (9), the following formula for calculating the manipulated variable MV is derived.

[0045]

number

[0046] Here, the PFC structure is illustrated with h=1, as shown in the block diagram in Figure 3. Reference orbital y in Figure 3 r The equation (10) is a calculation expression that combines the block and the controller block, and the manipulated variable MV is calculated using equation (10).

[0047] In Figure 3, A m B m This represents a discrete-time model. Here, we are using a model that includes a first-order lag and dead time, so it is as follows:

[0048]

number

[0049]

number

[0050] a1 and b0 are coefficients. Then, the second process model, which does not include dead time, can be expressed by the following equation (13).

[0051]

number

[0052] On the other hand, the first process model, which includes dead time, can be expressed by the following equation (14).

[0053]

number

[0054] When we set h=1 and correlate equation (3) with equation (13), the coefficients a1 and b0 of the discrete model are as follows.

[0055]

number

[0056] DCS5 calculates the manipulated variable MV using the PFC method described above and controls the controlled object so that the controlled variable PV asymptotically approaches the set value SV. In this embodiment, the control performance evaluation device 1 acquires operating data of the controlled object controlled by the PFC method from the plant information management device 2 and evaluates the PFC control performance index I c , and the model performance index I of the process model m This is used to calculate and evaluate the control performance of the PFC and the model performance of the process model.

[0057] The operating data is data measured or calculated for performing predictive control (PFC in this embodiment) that has a process model showing the dynamic characteristics of the controlled object. In this embodiment, the operating data is at least the predicted value PV of the control variable predicted using the process model. pred And, PV pred The trajectory value used to make the process follow the set value SV, the manipulated variable MV, and the output value y obtained by inputting the manipulated variable MV into the first process model. h This includes the following. The "orbital value" is the set value of orbital y described later. s Use this.

[0058] First, control performance index I c This will be explained. Figure 4 shows the control performance index I. c This is an explanatory diagram. The left side of Figure 4 shows the case where control performance is evaluated based on the difference (deviation) between the set value SV and the controlled variable PV, and the right side of Figure 4 shows the set value trajectory y s and the predicted value of the controlled variable PV PV pred The diagrams conceptually illustrate how control performance can be evaluated based on the difference between the two values.

[0059] As a method for evaluating control performance, an evaluation method that uses minimum variance control as a benchmark is known. Generally, when evaluating control performance, it is often evaluated based on the difference e between the setpoint SV and the controlled variable PV. However, there is a problem that if the setpoint SV is changed frequently, the control performance will be underestimated. Therefore, in this embodiment, the setpoint trajectory y s By introducing the set value trajectory y s and predicted PV pred This problem is avoided by evaluating control performance based on the difference e between the two values. Furthermore, since predicted values ​​are used in both cases, performance can be evaluated with greater accuracy than when using current values.

[0060] Specifically, for the purpose of evaluating control performance, the set value trajectory y is expressed by the following formula (16). s We introduce the following equation (16): The set value trajectory y at time h at time k is given by the equation (16). s It represents.

[0061]

number

[0062] If h=1, equation (16) can be expressed as equation (17) below.

[0063]

number

[0064] Here, the coefficient λ is the same as λ in equation (8), and the set value trajectory y s This is the sampling time T. s It can be calculated from the control parameter TRBF and the setpoint value SV.

[0065] In this embodiment, the set value trajectory y is determined according to the following formula (18). s and predicted PV pred Using the time-series data of the difference e, the control performance index I cThis calculates the following. Note that in control performance evaluation using minimum variance control as a benchmark, information on process dead time is required, but in the PFC of this embodiment, the predicted value PV pred This value is compensated for dead time and does not explicitly require information about process dead time, thus working more favorably in calculating the minimum variance.

[0066]

number

[0067] Here, σ 2 is the variance, and MV is e = y s -PV pred This represents the minimum variance of (note that MV in equation (18) is not a manipulated variable). If the control performance is good, I c It approaches 1, and if the control performance is poor, I c It approaches 0.

[0068] Thus, the set value trajectory y s By using the difference (deviation) e from the above, the control performance index I c This has the advantage of being less affected by changes in the set value. Furthermore, the set value trajectory y s The parameter λ and the reference orbit y r Since the parameter λ is the same, there is no need to set it again. Also, because the control parameter TRBF is a parameter that reflects the process delay, the control performance evaluation method also takes into account the ability to follow the set value.

[0069] The control performance evaluation device 1 predicts the value PV according to formula (18). pred and set value trajectory y s The ratio of the variance of the difference value e to the minimum variance of the difference value e is the control performance index I. c It is calculated as follows. The control performance evaluation device 1 uses the control performance index I c Determine whether the first threshold γ1 is greater than or equal to the control performance index I c If the value is greater than or equal to the first threshold γ1, a result indicating that there is no problem with the PFC's control performance is output to terminal 3.

[0070] In contrast, control performance index I c If the value is less than the first threshold γ1, i.e., if the control performance is low, the control performance evaluation device 1 evaluates whether the model performance is also low. Specifically, the control performance evaluation device 1 virtually remodels the process model (identifies the parameters of the new process model) and uses the data before and after remodeling to calculate the model performance index I m Calculate.

[0071] Figure 5 shows the model performance index I. m This is an explanatory diagram. The upper part of Figure 5 is a diagram extracted from the PFC block diagram exemplified in Figure 3, showing the parts corresponding to the process to be controlled and the first process model that takes dead time into consideration. The middle part of Figure 5 shows a third process model in which the parameters of the current process model have been virtually updated. The lower part of Figure 5 shows the model performance index I based on the output values ​​of the first process model and the output values ​​of the third process model. m The diagram shows how to calculate it.

[0072] If the control performance of the PFC is low, the control performance evaluation device 1 remodels the process model from the data of the manipulated variable MV and the controlled variable PV. In this embodiment, the ARX (Auto-Regressive eXogenous) model is used as the remodeling method. The ARX model is a discrete-time model represented by the following equations (19) to (21).

[0073]

number

[0074]

number

[0075]

number

[0076] y(t) is the controlled variable PV, and u(t) is the manipulated variable MV. In the present embodiment, in order to identify the process of the controlled object as a model including first-order lag and dead time, process identification is performed using time-series data of the manipulated variable MV and the controlled variable PV with the order of each variable set to first order, as shown in the following mathematical formulas (22) and (23).

[0077]

Math

[0078]

Math

[0079] Specifically, the control performance evaluation apparatus 1 identifies parameters a1 and b0 using data of the manipulated variable MV and the controlled variable PV. Here, from mathematical formula (15), as shown in the following mathematical formula (24), model gain K can be calculated from the identified parameters a1 and b0 m and time constant T m .

[0080]

Math

[0081] As described above, the control performance evaluation apparatus 1 identifies model gain K m and time constant T m such that the variance of error e m in FIG. 5 is minimized. More specifically regarding dead time, the control performance evaluation apparatus 1 shifts the dead time d defined in the process model by a predetermined value (one step at a time with d / Ts as the unit), and searches for model gain K m and time constant T m that minimize the variance of error e m . Accordingly, the control performance evaluation apparatus 1 obtains model gain K m and time constant T m that minimize the variance of error e mand the dead time d are identified as parameters of a new process model.

[0082] Note that the parameters a1 and b0 may be configured to be obtained sequentially using the recursive least squares method. In that case, the parameters a1 and b0 are obtained using the following mathematical expressions (25) to (29).

[0083] [[Math.]]

[0084] [[Math.]]

[0085] [[Math.]]

[0086] [[Math.]]

[0087] [[Math.]]

[0088] Here, ε represents the modeling error of the newly identified model.

[0089] The data of the controlled variable PV and the manipulated variable MV are data representing the dynamic characteristics of the controlled object, and identifying parameters using the ARX model from the data of the controlled variable PV and the manipulated variable MV is the same as transferring the dynamic characteristics of the controlled object to the process model in FIG. 5. Therefore, the error e before remodeling m and the error e' after virtual remodeling m can be used to evaluate the goodness of fit of the process model. Referring to FIG. 5, e m is the difference between the controlled variable PV and the output value y from the first process model hUse e m =PV-y h You can calculate it as follows, or refer to Figure 3 to get the predicted value PV with dead time compensation. pred and the output value y from the second process model m Use e m =PV pred -y m You can also calculate it that way.

[0090] The control performance evaluation device 1 calculates the error e' of the third process model after virtual remodeling. m And the error e of the first process model before remodeling m Based on this, Model Performance Index I m This calculates the model performance index I. m It is calculated according to the following formula (30).

[0091]

number

[0092] Here, in this embodiment, e m Herein are the controlled variable y and the output value y of the first process model. h This represents the error with e' m Herein are the controlled variable y and the output value y' of the third process model. h This represents the error with respect to the current process model. In other words, the control performance evaluation device 1 evaluates the error e in the current process model. m And the error e' in the new process model m The ratio to the model performance index I m It is calculated as e'. m Since the model is remodeled to minimize the variance, it follows the same approach as the control performance evaluation method described above. In model performance evaluation, if the model performance is good, I m It approaches 1, and if the model performance is poor, it becomes I m It approaches 0.

[0093] When successively determining parameters using methods such as successive least squares, the model performance index I is calculated successively using the modeling error ε as shown in equation (31). m You can configure it so that it is updated.

[0094]

number

[0095] Similarly, using the ARX model, the parameters of the first process model before remodeling and the setpoint trajectory y were obtained. s and reference orbit y r The parameters used in the calculation can be inversely calculated from the data. These parameters are control parameters set within the controller, but the effect is that they can be estimated from the operating data without referring to the control parameters.

[0096] Refer to Figure 3 for explanation. The input to the first process model is the manipulated variable MV, and its output is y h Therefore, change y(t) in equation (19) to y h (t), u(td / T s -1) to MV(td / T s -1) Using these data, from equations (22), (23), and (24), we obtain the parameter model gain K of the first process model. m and time constant T m This is estimated. Note that the dead time d is set to a predetermined value (d / T). s (Shifting by 1 step at a time using as the unit) the model gain K that minimizes the variance of the error. m and time constant T m Then, we identify the wasted time d.

[0097] On the other hand, the set value trajectory y s Since it is calculated using formula (17), similarly, y(t) in formula (19) is y s (t), u(td / T s Using -1) as the set value SV(t-1), and using these data, the time constant T can be found from equations (22), (23), and (24). sIdentify the time constant T from equation (8). s It is estimated that the control parameter TRBF is three times this value.

[0098] The control performance evaluation device 1 uses the model performance index I m Determine whether it is above the second threshold γ2. Model performance index I m If the value is greater than or equal to the second threshold γ2, i.e., if the model performance is high, it is determined that the cause of the low control performance is the influence of disturbances, and the determination result is output to terminal 3.

[0099] In contrast, Model Performance Index I m If the value is less than the second threshold γ2, i.e., if the model performance is low, it is determined that there is a mismatch between the process model and the actual process being controlled, and the determination result is output. At the same time, the control performance evaluation device 1 also identifies the model gain K as the third process model as a recommended parameter to be set in the new process model. m , time constant T m The time, including the dead time d, is also output to terminal 3. For reference, the parameters of the first process model, i.e., the parameters of the current process model, are used, specifically the model gain K. m , time constant T m The estimated values ​​of the dead time d and the control parameter TRBF may also be output to terminal 3.

[0100] Figure 6 shows an example of control performance evaluation results. Figure 6 shows an example of the evaluation results from the control performance evaluation device 1, as output to the screen of terminal 3. The example in Figure 5 shows a case where control performance deteriorates due to a model mismatch. Following the tag name and tag comment, the overall evaluation result and control performance index I are displayed. c Model performance index I m The estimated parameters of the current process model (first process model) and the recommended parameters (parameters of the third process model) are displayed on the screen of terminal 3.

[0101] Figure 7 shows an example of a trend graph. Figure 7 illustrates the first to fourth graphs output along with the control performance evaluation results in Figure 6. The first graph from the top shows the setpoint SV and the setpoint trajectory y. s The graphs show the trends (time series changes) of the controlled variable PV. The second graph in the second row shows the trend of the manipulated variable MV. The third graph in the third row shows the control performance index I. c Regarding this, the denominator is the set value trajectory y s and predicted PV pred This graph shows the trend of the difference e and the trend of the minimum variance of the difference e, which is the numerator. The fourth graph in the fourth row shows the model performance index I. m Regarding this, the error e between the denominator, the controlled variable PV, and the output value of the first process model (process model before remodeling) is... m The trend and the error e' between the numerator, the controlled variable PV, and the output value of the third process model (the remodeled process model). m This graph shows the trend.

[0102] Thus, the control performance evaluation device 1 performs a two-stage evaluation: first evaluating control performance, and then evaluating model performance. In the first stage of control performance evaluation, devices with low control performance due to the influence of disturbances are identified, and in the second stage of model performance evaluation, devices whose control performance can be improved by modifying the process model are identified. By using this two-stage evaluation, it is possible to distinguish between cases where low control performance is caused by disturbances and cases where low control performance is caused by model mismatch, and appropriate countermeasures can be taken for each case.

[0103] Furthermore, the control performance evaluation device 1 may perform the control performance evaluation after the model performance evaluation, rather than performing the control performance evaluation first.

[0104] Figure 8 is a flowchart showing an example of a processing procedure performed by the control performance evaluation device 1. Based on Figure 8, the processing content performed by the control performance evaluation device 1 will be explained.

[0105] The control unit 11 of the control performance evaluation device 1 acquires measured or calculated operating data from the plant information management device 2 in order to perform predictive control (PFC in this embodiment) using a first process model that shows the dynamic characteristics of the controlled object (step S11). The operating data includes at least the predicted value PV of the controlled quantity PV of the controlled object predicted using the first process model. pred And the trajectory value y to follow the set value SV. s And the manipulated variable MV and the output value y obtained by inputting the manipulated variable MV into the first process model. h Includes. Predicted PV pred This is the first output value y obtained by inputting the manipulated variable MV into the first process model that takes dead time into account from the controlled variable PV. h Subtracting this, and then inputting the manipulated variable MV into a second process model that does not include dead time, the second output value y is obtained. m This is the value after adding the dead time compensation. The orbital value is calculated using the PFC sampling time T. s The setpoint trajectory y is calculated from the control parameter TRBF and the setpoint SV. s That is the case.

[0106] The control unit 11 predicts the value of the controlled variable PV. pred The variance of the manipulated variable MV is calculated, and the predicted value PV is obtained. pred Step S12 determines whether the variance of is greater than or equal to the third threshold γ3, and whether the variance of the manipulated variable MV is greater than or equal to the fourth threshold γ4. If either is below the threshold (S12: NO), the determination result indicating that there is no problem is output to terminal 3 (Step S15), and the series of processes ends.

[0107] Predicted PV pred If the control unit 11 determines that the variance of is greater than or equal to the third threshold γ3 and the variance of the manipulated variable MV is greater than or equal to the fourth threshold γ4 (S12: YES), the control unit 11 sets the predicted value of the controlled variable PV PV pred And the set value trajectory y s Based on this, control performance index I for evaluating the control performance of PFC c The control unit calculates the predicted value PV (step S13). Specifically, the control unit 11 calculates the predicted value PV pred and set value trajectory ys The ratio of the variance of the difference value (deviation) e to the minimum variance of said difference value e is the control performance index I. c It is calculated as follows.

[0108] The control unit 11 controls the control performance index I c It is determined whether or not it is greater than or equal to the first threshold γ1 (step S14). Control performance index I c If it is determined that the value is greater than or equal to the first threshold γ1 (S14: YES), the control unit 11 outputs a determination result indicating that there is no problem with the control performance to the terminal 3 (step S15), and terminates the series of processes.

[0109] Control performance index I c If it is determined that the value is less than the first threshold γ1 (S14:NO), the control unit 11 performs a virtual remodeling of the first process model and identifies the parameters of a new third process model (step S16). Specifically, the control unit 11 determines the model gain K from the manipulated variable MV and the controlled variable PV. m and time constant T m Identify the error e. More specifically, the control unit 11 shifts the dead time d defined in the process model by a predetermined amount (one step at a time in units of d / Ts) while determining the error e. m By searching for the parameter that minimizes the variance of the error e, m The model gain K that minimizes the variance m and time constant T m Then, we identify the wasted time d.

[0110] The control unit 11 controls the controlled variable PV and the output value y' of the third process model (a new process model with identified parameters). h And the controlled variable PV and the output value y of the first process model (the current process model). h Based on this, a model performance index I for evaluating the performance of process models is used. m The control unit 11 calculates the error e' in the third process model (step S17). Specifically, the control unit 11 calculates the error e' in the third process model. m The variance and the error e in the first process model m The ratio of the variance to the model performance index I m It is calculated as follows.

[0111] The control unit 11 controls the model performance index I m Determine whether it is greater than or equal to the second threshold γ2 (Step S18). Model performance index I m If it is determined that the second threshold γ2 is greater than or equal to (S18: YES), the control unit 11 sets the control performance index I c The determination result that the low value is due to the influence of external disturbances is output to terminal 3 (step S19), and the series of processes ends.

[0112] Model Performance Index I m If it is determined that the control is less than the second threshold γ2 (S18:NO), the control unit 11 sets the control performance index I c and Model Performance Index I m The system outputs a determination result to terminal 3 indicating that the low value is due to a mismatch in the process model, and also outputs the parameters of the third process model identified in step S16 to terminal 3 as recommended values ​​(step S20), thus ending the series of processes.

[0113] Based on the above, according to this embodiment, it is possible to calculate a dimensionless index of control performance related to predictive control having a process model that shows the dynamic characteristics of the controlled object, regardless of the magnitude of the values ​​of the controlled variable and manipulated variable, and to suggest recommended model parameters for cases where the control performance is low due to a model mismatch.

[0114] The embodiments disclosed herein should be considered in all respects to be illustrative and not restrictive. The scope of the present invention is indicated by the claims, not in the sense described above, and all modifications within the sense and scope equivalent to the claims are intended.

[0115] The matters described in each embodiment can be combined with each other. Furthermore, the independent and dependent claims described in the claims can be combined with each other in any combination, regardless of the form of reference. In addition, the claims use a form in which claims referencing two or more other claims (multi-claim form), but are not limited to this. A form in which multi-claims referencing at least one multi-claim (multi-multi-claim) may also be used.

[0116] The contents of this disclosure are as follows:

[0117] (Note 1) For predictive control using a process model that shows the dynamic characteristics of the controlled object, data is obtained that includes a predicted value of the controlled quantity of the controlled object predicted using the process model that shows the dynamic characteristics of the controlled object, and a trajectory value for tracking a set value. Based on the predicted values ​​and trajectory values, the control performance index is calculated. A method for evaluating the control performance of a computer that performs processing. (Note 2) The ratio of the variance of the difference between the predicted value and the trajectory value to the minimum variance of the difference is calculated as the control performance index. The control performance evaluation method described in Appendix 1. (Note 3) The aforementioned predicted value is a time-compensated predicted value obtained by subtracting a first output value, which is obtained by inputting an manipulated variable to the controlled object, into a first process model that takes dead time into account, from the controlled variable, and then adding a second output value, which is obtained by inputting the manipulated variable into a second process model that does not include dead time. The control performance evaluation method described in Appendix 1 or 2. (Note 4) The trajectory value is a setpoint trajectory calculated from the sampling time of the predictive control, the control parameters, and the setpoint. A control performance evaluation method described in any of the appendices 1 to 3. (Note 5) The data further includes the manipulated variable for the controlled object, the output value obtained by inputting the manipulated variable into the process model, and the controlled variable for the controlled object. From the data of the manipulated and controlled variables, the parameters of the process model are identified. Based on the controlled variable and the output value of the newly identified process model, and the controlled variable and the output value of the current process model, a model performance index is calculated. A control performance evaluation method described in any of the appendices 1 to 4. (Note 6) The ratio of the variance of the error between the controlled variable and the output value of the new process model to the variance of the error between the controlled variable and the output value of the current process model is calculated as the model performance index. The control performance evaluation method described in Appendix 5. (Note 7) If the model performance indicator is below the threshold, the identified parameter is output as a recommended value. The control performance evaluation method described in Appendix 5 or 6. (Note 8) By searching for the parameter that minimizes the variance of the error between the controlled variable and the output value of the process model while shifting the dead time defined in the process model by a predetermined amount, the parameter and the dead time are identified. A control performance evaluation method described in any of the appendices 5 to 7. (Note 9) If the control performance index is equal to or greater than the first threshold, it is determined that there is no problem with the control performance. If the control performance index is below the first threshold and the model performance index is above the second threshold, it is determined that the cause of the low control performance index is the influence of disturbances. If the control performance index is below the first threshold and the model performance index is below the second threshold, it is determined that the reason for the low control performance index and model performance index is a mismatch in the process model. Output the judgment result. A control performance evaluation method described in any of the appendices 5 to 8. (Note 10) Predictive control (PFC) is a process model that shows the dynamic characteristics of the controlled object. A control performance evaluation method described in any of the appendices 1 to 9. (Note 11) For predictive control using a process model that shows the dynamic characteristics of the controlled object, data is obtained that includes a predicted value of the controlled quantity of the controlled object predicted using the process model that shows the dynamic characteristics of the controlled object, and a trajectory value for tracking a set value. Based on the predicted values ​​and trajectory values, the control performance index is calculated. A control performance evaluation program that has a computer execute a process. (Note 12) A control performance evaluation device comprising a control unit, The control unit, For predictive control using a process model that shows the dynamic characteristics of the controlled object, data is obtained that includes a predicted value of the controlled quantity of the controlled object predicted using the process model that shows the dynamic characteristics of the controlled object, and a trajectory value for tracking a set value. Based on the predicted values ​​and trajectory values, the control performance index is calculated. Control performance evaluation device. (Note 13) Data is obtained for predictive control using a process model that shows the dynamic characteristics of a controlled object, and includes a predicted value of a control quantity predicted using the process model that shows the dynamic characteristics of the controlled object, an manipulated quantity for the controlled object, an output value obtained by inputting the manipulated quantity into the process model, and the control quantity of the controlled object. From the data of the manipulated and controlled variables, the parameters of the process model are identified. Based on the predicted values ​​and the output values ​​of the newly identified process model, and the predicted values ​​and the output values ​​of the current process model, a model performance index is calculated. A model performance evaluation method in which a computer performs the processing. (Note 14) A control performance index is obtained from the operating data of the controlled object, which is predictively controlled using a first process model that shows the dynamic characteristics of the controlled object. A third process model is obtained by identifying parameters from the operating data of the controlled object, the first process model is obtained, and a model performance index is calculated using the operating data. If the control performance index is below the first threshold and the model performance index is below the second threshold, the control performance index and the model performance index are output, and the parameters of the third process model are output as recommended parameters. A program that instructs a computer to perform a process. (Note 15) The control performance index is calculated from the difference data between the predicted value of the controlled quantity of the controlled object predicted using the first process model and the trajectory value used to make the predicted value follow the set value. The program described in Appendix 14. (Note 16) The ratio of the variance of the difference between the predicted value and the trajectory value to the minimum variance of the difference is calculated as the control performance index. The program described in Appendix 15. (Note 17) A third graph showing the trend of the difference between the predicted value and the trajectory value, and the trend of the minimum variance of the difference value, is displayed on the display unit. The program described in Appendix 16. (Note 18) Using the set value data and the orbital value data, the control parameters used to calculate the orbital value are identified. The program described in any of the appendices 15-17. (Note 19) The third process model is identified from the manipulated and controlled variable data included in the operation data, The aforementioned model performance index is calculated from the difference data between the output value and the controlled variable of the third process model and the difference data between the output value and the controlled variable of the first process model. The program described in any of the appendices 14-18. (Note 20) The ratio of the variance of the error between the output value and the controlled variable of the third process model to the variance of the error between the output value and the controlled variable of the first process model is calculated as the model performance index. The program described in Appendix 19. (Note 21) A fourth graph showing the error trend between the output value and the controlled variable of the third process model, and the error trend between the output value and the controlled variable of the first process model, is displayed on the display unit. The program described in Appendix 20. (Note 22) When the control performance index is equal to or greater than the first threshold, the control performance index and information indicating that there are no problems with the control performance are output. The program described in any of the appendices 14-21. (Note 23) If the control performance index is below the first threshold and the model performance index is above the second threshold, the control performance index and the model performance index, along with information indicating that the decrease in control performance is due to disturbances, are output. The program described in any of the appendices 14-22. (Note 24) If the control performance index is below the first threshold and the model performance index is below the second threshold, the control performance index, the model performance index, the parameters of the third process model, and information indicating that the parameters of the process model need to be modified are output. The program described in any of the appendices 14-23. (Note 25) If the variance of the predicted value of the controlled variable of the controlled object, or the variance of the controlled variable, predicted using the first process model, is less than the third threshold, or if the variance of the manipulated variable of the controlled object is less than the fourth threshold, information indicating that there are no problems with the control performance is output, regardless of the control performance index and the model performance index. The program described in any of the appendices 14-24. (Note 26) The parameters of the first process model are identified using data consisting of an operation variable applied to the controlled object and an output value calculated by inputting the operation variable into the first process model, which includes dead time. The program described in any of the appendices 14-25. (Note 27) A control performance index is obtained from the operating data of the controlled object, which is predictively controlled using a first process model that shows the dynamic characteristics of the controlled object. A third process model is obtained by identifying parameters from the operating data of the controlled object, the first process model is obtained, and a model performance index is calculated using the operating data. If the control performance index is below the first threshold and the model performance index is below the second threshold, the control performance index and the model performance index are output, and the parameters of the third process model are output as recommended parameters. A method for evaluating the control performance of a computer that performs processing. (Note 28) A control performance evaluation device comprising a control unit, The control unit, A control performance index is obtained from the operating data of the controlled object, which is predictively controlled using a first process model that shows the dynamic characteristics of the controlled object. A third process model is obtained by identifying parameters from the operating data of the controlled object, the first process model is obtained, and a model performance index is calculated using the operating data. If the control performance index is below the first threshold and the model performance index is below the second threshold, the control performance index and the model performance index are output, and the parameters of the third process model are output as recommended parameters. Control performance evaluation device. [Explanation of Symbols]

[0118] 1. Control performance evaluation device 11 Control Unit 12 Main memory 13 Communications Department 14 Auxiliary storage P Program 2. Plant Information Management System 3 terminals 4 OPC Servers 5 DCS

Claims

1. For predictive control using a process model that shows the dynamic characteristics of the controlled object, data is obtained that includes a predicted value of the controlled quantity of the controlled object predicted using the process model that shows the dynamic characteristics of the controlled object, and a trajectory value for tracking a set value. Based on the predicted values ​​and trajectory values, the control performance index is calculated. A method for evaluating the control performance of a computer that performs processing, The ratio of the variance of the difference between the predicted value and the trajectory value to the minimum variance of the difference is calculated as the control performance index. A method for evaluating the control performance of a computer that performs processing.

2. Data measured or calculated for predictive control using a process model that shows the dynamic characteristics of a controlled object, the data includes a predicted value of the controlled quantity of the controlled object predicted using the process model that shows the dynamic characteristics of the controlled object, and a trajectory value for following a set value, Based on the predicted values ​​and trajectory values, the control performance index is calculated. A method for evaluating the control performance of a computer that performs processing, The data further includes the manipulated variable for the controlled object, the output value obtained by inputting the manipulated variable into the process model, and the controlled variable for the controlled object. From the data of the manipulated and controlled variables, the parameters of the process model are identified. The ratio of the variance of the error between the controlled variable and the output value of the newly identified process model to the variance of the error between the controlled variable and the output value of the current process model is calculated as a model performance index. A method for evaluating the control performance of a computer that performs processing.

3. Data measured or calculated for predictive control using a process model that shows the dynamic characteristics of a controlled object, the data includes a predicted value of the controlled quantity of the controlled object predicted using the process model that shows the dynamic characteristics of the controlled object, and a trajectory value for tracking a set value, Based on the predicted values ​​and trajectory values, the control performance index is calculated. A method for evaluating the control performance of a computer that performs processing, The data further includes the manipulated variable for the controlled object, the output value obtained by inputting the manipulated variable into the process model, and the controlled variable for the controlled object. From the data of the manipulated and controlled variables, the parameters of the process model are identified. Based on the controlled variable and the output value of the newly identified process model, and the controlled variable and the output value of the current process model, a model performance index is calculated. By searching for the parameter that minimizes the variance of the error between the controlled variable and the output value of the process model while shifting the dead time defined in the process model by a predetermined amount, the parameter and the dead time are identified. A method for evaluating the control performance of a computer that performs processing.

4. The aforementioned predicted value is a time-compensated predicted value obtained by subtracting a first output value, which is obtained by inputting an manipulated variable to the controlled object, into a first process model that takes dead time into account from the controlled variable, and then adding a second output value, which is obtained by inputting the manipulated variable into a second process model that does not include dead time. A method for evaluating control performance according to any one of claims 1 to 3.

5. The trajectory value is a setpoint trajectory calculated from the sampling time of the predictive control, the control parameters, and the setpoint. A method for evaluating control performance according to any one of claims 1 to 3.

6. If the model performance indicator is below the threshold, the identified parameter is output as a recommended value. The control performance evaluation method according to claim 2 or 3.

7. If the control performance index is equal to or greater than the first threshold, it is determined that there is no problem with the control performance. If the control performance index is below the first threshold and the model performance index is above the second threshold, it is determined that the cause of the low control performance index is the influence of disturbances. If the control performance index is below a first threshold and the model performance index is below a second threshold, it is determined that the reason for the low control performance index and model performance index is a mismatch in the process model. Output the judgment result. The control performance evaluation method according to claim 2 or 3.

8. Predictive control (PFC) is a predictive control that has a process model showing the dynamic characteristics of the controlled object. A method for evaluating control performance according to any one of claims 1 to 3.

9. For predictive control using a process model that shows the dynamic characteristics of the controlled object, data is obtained that includes a predicted value of the controlled quantity of the controlled object predicted using the process model that shows the dynamic characteristics of the controlled object, and a trajectory value for tracking a set value. Based on the predicted values ​​and trajectory values, the control performance index is calculated. A control performance evaluation program that causes a computer to execute a process, The ratio of the variance of the difference between the predicted value and the trajectory value to the minimum variance of the difference is calculated as the control performance index. A control performance evaluation program that has a computer execute a process.

10. Data measured or calculated for predictive control using a process model that shows the dynamic characteristics of a controlled object, the data includes a predicted value of the controlled quantity of the controlled object predicted using the process model that shows the dynamic characteristics of the controlled object, and a trajectory value for tracking a set value, Based on the predicted values ​​and trajectory values, the control performance index is calculated. A control performance evaluation program that causes a computer to execute a process, The data further includes the manipulated variable for the controlled object, the output value obtained by inputting the manipulated variable into the process model, and the controlled variable for the controlled object. From the data of the manipulated and controlled variables, the parameters of the process model are identified. The ratio of the variance of the error between the controlled variable and the output value of the newly identified process model to the variance of the error between the controlled variable and the output value of the current process model is calculated as a model performance index. A control performance evaluation program that has a computer execute a process.

11. Data measured or calculated for predictive control using a process model that shows the dynamic characteristics of a controlled object, the data includes a predicted value of the controlled quantity of the controlled object predicted using the process model that shows the dynamic characteristics of the controlled object, and a trajectory value for tracking a set value, Based on the predicted values ​​and trajectory values, the control performance index is calculated. A control performance evaluation program that causes a computer to execute a process, The data further includes the manipulated variable for the controlled object, the output value obtained by inputting the manipulated variable into the process model, and the controlled variable for the controlled object. From the data of the manipulated and controlled variables, the parameters of the process model are identified. Based on the controlled variable and the output value of the newly identified process model, and the controlled variable and the output value of the current process model, a model performance index is calculated. By searching for the parameter that minimizes the variance of the error between the controlled variable and the output value of the process model while shifting the dead time defined in the process model by a predetermined amount, the parameter and the dead time are identified. A control performance evaluation program that has a computer execute a process.

12. A control performance evaluation device comprising a control unit, The control unit, For predictive control using a process model that shows the dynamic characteristics of the controlled object, data is obtained that includes a predicted value of the controlled quantity of the controlled object predicted using the process model that shows the dynamic characteristics of the controlled object, and a trajectory value for tracking a set value. Based on the predicted values ​​and trajectory values, the control performance index is calculated. A control performance evaluation device, The control unit calculates the ratio of the variance of the difference between the predicted value and the trajectory value to the minimum variance of the difference as the control performance index. Control performance evaluation device.

13. A control performance evaluation device comprising a control unit, The control unit, For predictive control using a process model that shows the dynamic characteristics of the controlled object, data is obtained that includes a predicted value of the controlled quantity of the controlled object predicted using the process model that shows the dynamic characteristics of the controlled object, and a trajectory value for tracking a set value. Based on the predicted values ​​and trajectory values, the control performance index is calculated. A control performance evaluation device The data further includes the manipulated variable for the controlled object, the output value obtained by inputting the manipulated variable into the process model, and the controlled variable for the controlled object. The control unit, From the data of the manipulated and controlled variables, the parameters of the process model are identified. The ratio of the variance of the error between the controlled variable and the output value of the newly identified process model to the variance of the error between the controlled variable and the output value of the current process model is calculated as a model performance index. Control performance evaluation device.

14. A control performance evaluation device comprising a control unit, The control unit, For predictive control using a process model that shows the dynamic characteristics of the controlled object, data is obtained that includes a predicted value of the controlled quantity of the controlled object predicted using the process model that shows the dynamic characteristics of the controlled object, and a trajectory value for tracking a set value. Based on the predicted values ​​and trajectory values, the control performance index is calculated. A control performance evaluation device, The data further includes the manipulated variable for the controlled object, the output value obtained by inputting the manipulated variable into the process model, and the controlled variable for the controlled object. The control unit, From the data of the manipulated and controlled variables, the parameters of the process model are identified. Based on the controlled variable and the output value of the newly identified process model, and the controlled variable and the output value of the current process model, a model performance index is calculated. By searching for the parameter that minimizes the variance of the error between the controlled variable and the output value of the process model while shifting the dead time defined in the process model by a predetermined amount, the parameter and the dead time are identified. Control performance evaluation device.

15. Data is obtained for predictive control using a process model that shows the dynamic characteristics of a controlled object, and includes a predicted value of a control quantity predicted using the process model that shows the dynamic characteristics of the controlled object, an manipulated quantity for the controlled object, an output value obtained by inputting the manipulated quantity into the process model, and the control quantity of the controlled object. From the data of the manipulated and controlled variables, the parameters of the process model are identified. Based on the predicted values ​​and the output values ​​of the newly identified process model, and the predicted values ​​and the output values ​​of the current process model, a model performance index is calculated. A model performance evaluation method in which a computer performs the processing, The ratio of the variance of the error between the predicted value and the output value of the new process model to the variance of the error between the predicted value and the output value of the current process model is calculated as the model performance index. A model performance evaluation method in which a computer performs the processing.

16. Data obtained for predictive control using a process model that shows the dynamic characteristics of a controlled object, the data includes a predicted value of a control quantity predicted using the process model that shows the dynamic characteristics of the controlled object, an manipulated quantity for the controlled object, an output value obtained by inputting the manipulated quantity into the process model, and the control quantity of the controlled object. From the data of the manipulated and controlled variables, the parameters of the process model are identified. Based on the predicted values ​​and the output values ​​of the newly identified process model, and the predicted values ​​and the output values ​​of the current process model, a model performance index is calculated. A model performance evaluation method in which a computer performs the processing, By shifting the dead time defined in the process model by a predetermined amount and searching for the parameter that minimizes the variance of the error between the predicted value and the output value of the process model, the parameter and the dead time are identified. A model performance evaluation method in which a computer performs the processing.

17. If the model performance index is below a threshold, the identified parameter is output as a recommended value. The model performance evaluation method according to claim 15 or 16.

Images