Process control device and process control method

Abstract

To provide a process control device with high stability and follow-up to control target values.SOLUTION: The process control device for calculating an operation amount command value of an operation end on the basis of measured values at a measuring end, includes: a state transition probability model that quantifies the probability of a certain process state transiting to any process state after a predetermined time cycle; a next target state calculation unit that calculates a target state to be aimed at in a next control cycle on the basis of the state transition probability model; and a command value calculation unit that calculates the operation amount command value that achieves the target state and supplies it to the operation end. The next target state calculation unit is configured to input a current process state into the state transition probability model to calculate a transition destination candidate, determine whether the transition destination candidate lies between the current process state and a final target process state, input the transition destination candidate into the state transition probability model to recalculate the transition destination candidate if a determination result is false, and set the transition destination candidate to the target state if the determination result is true.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a process control device and a process control method for controlling a process system such as a plant in real time. [Background technology]

[0002] In recent years, methods that apply machine learning technology have been proposed as real-time control methods for process systems such as plants. One such method uses a state transition probability model that represents the probability with which a process state will transition to another process state after a specified time.

[0003] For example, paragraphs 0033, 0067 to 0074 of Patent Document 1 describe that, for each source state si, among the destination states sj that can be transitioned from the source state si, a state sj* that is easiest to transition to the target state sgoal is calculated, the state with the highest value of the destination state value function V is determined as the destination state to be targeted in the next control cycle, and the operation amount a required to transition to this targeted state is calculated. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 2019-159876 Summary of the Invention [Problem to be solved by the invention]

[0005] When applying machine learning to real-time control of process systems such as plants, the following issues must be taken into consideration.

[0006] <First issue> The first problem is that the response of the measurement value of the measurement terminal of the controlled object to the operation of the control terminal of the controlled object is delayed.

[0007] When the response delay is significant, there is a large difference between the timing of a control operation and the timing when its effect appears in the controlled variable of the controlled object. If time-series data with such a timing difference is subjected to machine learning as is without any special ingenuity, even if a causal relationship exists in the actual process, that is, "a control operation is performed, and the controlled variable changes as a result," the model acquired through machine learning will tend to learn the relationship between the two as if they are uncorrelated due to the phase difference between the control operation (change in the controlled variable) and its result (change in the controlled variable), making it difficult to learn the plant characteristics with sufficient accuracy for use in control.

[0008] To address problems caused by response delays, it is necessary to devise learning and control methods that take into account the characteristics of response delays.Response delays can be broadly divided into two types: dead time elements, in which no response appears for a certain period of time, and first-order lag elements, in which a response appears immediately after operation but it takes time for the entire response to appear, without any period of no response.

[0009] As a measure against dead time, for example, when a control operation is executed, it is assumed that there will be a delay corresponding to the dead time before the operation shows its effect on the controlled variable, and instead of executing the control operation based on the current measured value of the controlled variable, it is necessary to execute the control operation based on a predicted value of the controlled variable that is a distance into the future corresponding to the dead time estimated by some kind of predictive calculation (the simplest case being a regression equation obtained inductively in advance).Furthermore, as a measure against delay elements such as first-order lag, it is necessary to devise a method for learning data and calculating the execution method of the control operation, taking into account the type of delay element and the phase of the waveform according to the parameter value.

[0010] As such, effective countermeasures against response delays vary depending on the type and nature of the delay. However, in actual plant control, response delays often appear as a mixture of various types of delays. This makes it difficult to take systematic, consistent countermeasures, and it is necessary to consider individual control methods depending on the nature of the delay in the target process, which requires a great deal of effort to consider and establish a control method.

[0011] <Second issue> The second problem is that there is often a process in which a large disturbance occurs between the measurement end of the controlled variable to be controlled and the operation end that executes the control operation.

[0012] When disturbances are large and irregular, it becomes difficult to learn plant characteristics, such as differences in plant state depending on operating conditions and the relationship between appropriate control operations and process input fluctuations, with sufficient accuracy for use in control through machine learning. This issue becomes even more serious under the first issue mentioned above (conditions where there is a large response delay to control operations).

[0013] <Third issue> The third challenge is the difficulty of applying machine learning when the control of the target process interferes with the control of other processes.

[0014] Plant process control systems often have complex hierarchical structures, with individual controls for each unit process indirectly connected through the process flow. In addition to these individual unit process controls, there are also controls spanning multiple unit processes and plant-wide controls. Such control systems often have complex interference loops where the results of one unit process affect the control of another unit process, another higher-level system, or the entire plant. Plant control using existing methods (classical control theory and modern control theory) addresses this interference by tuning the control parameters of each process to prevent problems from becoming too apparent, resulting in practically problem-free operation. However, when introducing machine learning-based control to specific parts of such existing plant control systems, it is difficult to conduct learning that takes into account the mutual influence and interference with other control systems in advance.

[0015] The reason why learning such an interference system is difficult is as follows.

[0016] First, because the behavior of existing control systems has not yet manifested to the extent that interference becomes a problem, it is often not clear in advance how far the learning range should be expanded to anticipate interference, or how many measurement signals should be learned.Since these are problems that do not occur during normal plant operation, it is often difficult to specifically recognize them as potential problems.

[0017] Second, machine learning of the behavior of an interference system requires the inclusion of multiple control systems and their interactions in the training data, which requires a huge amount of computer memory capacity. In reality, this capacity is often insufficient. Often, the target data to be trained cannot include all information about other interfering control systems, making it necessary to narrow the training data to focus on those related to the behavior of the controlled object. When implementing control using such narrowly-scoped training data, the control behavior required to achieve higher-precision control often becomes more aggressive, with larger or steeper swings than conventional control. This increases control interference and can lead to instability, such as large fluctuations in the state of the target process. Therefore, the control behavior must be more restrictive. However, no efficient method has been clarified yet for identifying the desired restrictive control behavior.

[0018] <Relationship between the above problem and Patent Document 1> However, Patent Document 1 merely proposes a method for calculating a target state in the next control cycle using a state transition probability model, and does not disclose any specific solutions to the above three problems.

[0019] Therefore, an object of the present invention is to provide a process control device and a process control method that can achieve stable control with high tracking ability to a control target value for processes that are difficult to control, such as when there is a large delay until a change in the controlled variable to be controlled appears in response to the control operation of the process, when there is a large disturbance occurring between the control operation terminal and the measurement terminal of the controlled variable, or when control of the target process interferes with control of other processes in the plant. [Means for solving the problem]

[0020] In order to solve the above problems, the present invention provides a process control device that controls a process system equipped with a control element and a measurement element, and calculates a manipulated variable command value for the control element based on a measurement value of the measurement element, the process control device comprising: a state transition probability model that quantifies the probability of a process state to which a certain process state will transition after a predetermined time period; a next target state calculation unit that calculates a target state to be aimed for in the next control period based on the state transition probability model; and a command value calculation unit that calculates the manipulated variable command value that achieves the target state and supplies the calculated manipulated variable command value to the control element, wherein the next target state calculation unit inputs a current process state to the state transition probability model to calculate transition destination candidates, determines whether the transition destination candidates are between the current process state and a final target process state, and if the determination result is false, inputs the transition destination candidates into the state transition probability model to recalculate the transition destination candidates, and if the determination result is true, sets the transition destination candidates to the target state. [Effects of the Invention]

[0021] The process control device and process control method of the present invention can achieve stable control with high tracking ability to the control target value for processes that are difficult to control, such as when there is a large delay until a change in the controlled variable to be controlled appears in response to the control operation of the process, when there is a large disturbance occurring between the control operation terminal and the controlled variable measurement terminal, or when the control of the target process interferes with the control of other processes in the plant. [Brief explanation of the drawings]

[0022] [Figure 1] Functional block diagram of a process control system according to a first embodiment [Figure 2] 1 is a flowchart of a calculation process of a manipulated variable command value executed in the first embodiment. [Figure 3] FIG. 10 is a schematic diagram illustrating an example of a time series transition of a process control amount when the first embodiment is applied. [Figure 4] 10 is a flowchart of a calculation process of a manipulated variable command value executed in the second embodiment. [Figure 5] 10 is an example of a setting screen for a process control device according to the second embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0023] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. [Example]

[0024] First, a process control system 100 according to a first embodiment will be described with reference to FIGS.

[0025] <Process Control System 100> 1 is a functional block diagram of a process control system 100 according to a first embodiment. As shown in the figure, the process control system 100 includes a process system 1 to be controlled and a process control device 2 that controls the process system 1 in real time.

[0026] The process system 1 is a system (e.g., a chemical plant, a steel plant, a power plant, etc.) that executes various processes (e.g., chemical product manufacturing, steelmaking, power generation, etc.), and has an operating end 11 and a measuring end 12. The operating end 11 is a valve or the like that is directly controlled by the process control device 2, and by operating this, it is possible to control the controlled variable of the process system 1 (e.g., temperature, pressure, flow rate, etc.). The measuring end 12 is a sensor installed in the process system 1, and outputs the measured value of the controlled variable of the process system 1 to the process control device 2. Note that while the process system 1 in FIG. 1 illustrates a configuration that includes one operating end 11 and one measuring end 12, multiple operating ends 11 and multiple measuring ends 12 may be provided depending on the configuration of the process system 1.

[0027] The process control device 2 is a device that controls the controlled variable of the process system 1 so that it is approximately equal to a desired control target value (e.g., a target temperature, a target pressure, a target flow rate, etc.) by controlling the controlled variable 11 based on the measurement value of the measurement terminal 12, and is equipped with a controlled variable calculation unit 21 and a learning model 22. Specifically, the process control device 2 is a computer equipped with hardware such as an arithmetic unit such as a CPU, a main storage device such as a semiconductor memory, an auxiliary storage device such as a hard disk, and a communication device. The arithmetic unit executes a predetermined program to realize functional units such as the controlled variable calculation unit 21, but the following will explain each unit in detail, omitting such well-known techniques.

[0028] The manipulated variable calculation unit 21 is a functional unit that calculates a manipulated variable command value to be given to the manipulated variable 11 based on the measurement values ​​of the measurement terminal 12, and includes a next target state calculation unit 21a and a command value calculation unit 21b. The next target state calculation unit 21a is a functional unit that calculates a post-transition process state S' (described later) based on the pre-transition process state S defined by the measurement values ​​and manipulated variable command values ​​of the current control cycle, and calculates the process state to be aimed for in the next control cycle (hereinafter referred to as the "next target state"). The command value calculation unit 21b is a functional unit that calculates a manipulated variable command value for the manipulated variable 11 that will achieve the next target state calculated by the next target state calculation unit 21a, and supplies this to the manipulated variable 11 prior to the next control cycle. If multiple manipulated variables 11 are provided, the command value calculation unit 21b can calculate the same number of manipulated variable command values ​​as the number of manipulated variables 11 so that each manipulated variable can be controlled individually.

[0029] Here, the process state S before the transition is a process state defined by the value of any one parameter among one or more manipulated variable command values ​​sent by the process control device 2 to the process system 1 (operating end 11) and / or one or more measurement values ​​received by the process control device 2 from the process system 1 (measuring end 12) in the current control cycle, or a process state defined by a combination of any multiple parameter values.

[0030] In this way, the process state can be defined by one method in which the process state is composed of one or more measurement values ​​and does not include a manipulated variable command value, or by one method in which a manipulated variable command value is included.

[0031] The method of not including manipulated variable command values ​​is suitable when the calculation method for the manipulated variable to realize the state transition defined by the measurement value is clear and there is a high probability that the state transition will actually be realized as the calculation result. By not including manipulated variable command values ​​in the state, the number of variables (dimensions) that define the state is reduced, and the number of states divided into meshes in the multidimensional state space described below can be kept small, thereby reducing memory consumption during calculation. In addition, because the number of meshes in the state is small, learning (described later in learning model 22) can be performed with sufficient accuracy even if the number of learning data is smaller (compared to when manipulated variable command values ​​are included).

[0032] The inclusion of manipulated variable command values ​​is suitable when it is desirable to learn the relationship between manipulated variable command values ​​and measured values. For example, when a measured value transitions from one value to another, the corresponding relationship between the value and the manipulated variable command value is highly reproducible, and the control characteristics can be effectively learned and modeled (described later in the learning model 22).

[0033] To explain more specifically the case where the process state S is defined by a combination of values ​​of multiple parameters, the position in a multidimensional space formed by combining a finite number of parameters and determined by the combination of the values ​​of each parameter represents the process state S. Typically, the process state S is defined as a multidimensional mesh in which each dimension of the process state is divided (discretized) into a finite number of regions and combined.

[0034] The learning model 22 is a trained model generated by machine learning the state transition performance of the process system 1, and includes a state transition probability model M that quantifies the probability that a certain process state S will transition to a certain process state S' after a specified time period.

[0035] <Details of how to calculate the manipulated variable command value> Next, the details of the calculation process of the manipulated variable command value, which is executed in each control cycle by the manipulated variable calculation unit 21, will be described with reference to the flowchart of FIG.

[0036] First, in step St1, the operation amount calculation unit 21 starts the calculation process of the operation command value, which is performed for each control period.

[0037] Next, in step St2, the next target state calculation unit 21a sets the current process state of the process system 1 to the process state S before the transition, which is the starting point for calculating the state transition.

[0038] In step St3, the next target state calculation unit 21a inputs the pre-transition process state S into the state transition probability model M to calculate the process state of the next cycle (post-transition process state S'). The post-transition process state S' calculated here is a candidate for the next target state, and will be referred to as a "transition destination candidate" hereinafter. Note that, since the transition destination candidate is merely a candidate for the next target state, it is not immediately used in the calculation of the manipulated variable command value, but can be updated by repeated calculations (recalculation again in step St3 after passing through steps St4 and St5) described below.

[0039] Here, the state transition probability model M typically represents which process state each of the discretized process states will transition to after a predetermined time period. For example, if there are N possible process states, and the state number of the process state S before the transition is i and the state number of the process state S' after the transition is j, the probability of transitioning from the process state Si to the process state S'j is P ij Alternatively, the state transition probability model M may be one in which the process states are defined in a continuous space that is not discretized, and the transition probability P(S, S') from the process state S before the transition to the process state S' after the transition is expressed as a continuous function.

[0040] Furthermore, while the above description of the state transition probability model M has been given assuming that one state transition in a certain time period is modeled, the state transition probability model M used in this embodiment may also be one that models a plurality of state transitions. As an example, a weighted average of the state transition probability models M for each of the number of state transitions, such as 1, 2, 3, ..., can be given. When a state transition probability matrix for one state transition is represented as P in the discretized state definition, such a state transition probability model M can be expressed as shown in Equation 1 using a damping coefficient γ that satisfies 0<γ<1.

[0041]

number

[0042] As another method, it is also possible to assume an infinite number of state transitions, rather than limiting the upper limit to a finite number, and use the weighted average of the state transition probability model M for each number of transitions as the state transition probability model M. In the case of a discretized state definition, this state transition probability model M is obtained by taking the limit N→∞ in the above (Equation 1), and by rearranging the equation and representing the unit matrix as I, it can be expressed as Equation 2.

[0043]

number

[0044] where (I-γP) -1 is the inverse matrix of the matrix (I-γP), and β is a constant coefficient, which can be set as, for example, β=1-γ.

[0045] The calculation of the transition destination candidates in step St3 using these state transition probability models M is performed by calculating and obtaining a process state that is estimated to be likely to transition to as the next process state and to have a high probability of subsequently reaching the final target state, based on the respective probabilities of transitioning from the current process state S input into the state transition probability model M to each possible next process state S' and the likelihood of reaching the final target state from each possible transition destination process state S', calculated or set in a predetermined procedure.

[0046] As a specific method, the calculation procedure shown in Patent Document 1 can be used. Typically, the state transition probability model M for one state transition is referenced to calculate the probability P of each transition from the current process state Si to each of the next possible process states S′j. ij is obtained, and the value of each row of the column (assuming it is the k-th column) representing the final target state in the matrix of (Equation 2) is multiplied as an evaluation value of the likelihood of reaching the final target state from each process state S'j that can be the transition destination. (This represents the average likelihood of transition from each process state S'j to the final target state Sk, taking into account all patterns of the number of state transitions.) The process state S'j with the largest multiplied value is set as the next target state.

[0047] In this example, the matrix of (Equation 2) is used to represent the likelihood of reaching the final goal state from each process state S'j that can be a transition destination, but the matrix of (Equation 1) may be used instead. The suitability of each is as follows:

[0048] The matrix expressed as in (Equation 1) corresponds to an average of the state transition model M for each of a finite number of state transitions, with the upper limit being the number of state transitions corresponding to a time scale, when fluctuations such as disturbances in the process input to the process system 1 and their effects continue for approximately the length of that time scale. Therefore, when the matrix in (Equation 1) is used, even if it is difficult for the process state to approach the final target state in one control cycle due to fluctuations such as disturbances, for a process system 1 having such phenomenon characteristics, it is possible to repeat state transitions by broadly assuming the possibility of reaching the final target state in multiple control cycles, thereby improving the stability of control performance.

[0049] On the other hand, the matrix expressed as in (Equation 2) assumes a wide range of state transition patterns, from one state transition to an infinite number of state transitions, and extracts the likelihood of the averaged state transitions. Therefore, when the matrix of (Equation 2) is used, the final target state can be effectively approached for a process system 1 in which any number of state transitions can be repeated until the process state reaches the final target state and it is important to reach the final target state, or for a process system 1 in which it is difficult to easily identify the time scale over which the effects of destabilizing factors such as disturbances remain. Even if the final target state is deviated from due to disturbances, the process state can be brought closer to the final target state more accurately and effectively.

[0050] In step St4, the next goal state calculation unit 21a determines whether the transition destination candidate calculated in step St3 is between the current process state S and the preset final goal state. If the transition destination candidate is not between them, the process proceeds to step St5 to update the transition destination candidate, and if the transition destination candidate is between them, the process proceeds to step St6 to set the transition destination candidate as the next goal state.

[0051] Here, the final target state is a process state in which, in a multidimensional space based on the definition above, at least the controlled variable value of a controlled parameter (e.g., temperature, pressure, flow rate, etc.) substantially matches a control target value. For example, when the process state is discretely defined, this can be defined as a state in which the control target value is included in a predetermined divided region. With such a definition, if the division width of the discretized controlled variable is sufficiently fine, it is possible to determine with sufficient accuracy for practical use whether the control variable matches the control target value or whether the magnitude is compared (i.e., whether the next target value of the controlled variable is between the current process state value and the control target value).

[0052] Furthermore, when the process state is not discretized but treated continuously, it is also possible to define the process state in which the control variable value of the controlled parameter is within a predetermined upper and lower limit range from the control target value as the final target state (in this case, by providing a range for the target state, stable control is expected to be achieved).

[0053] Furthermore, the final target state may be defined not only on the condition that the control quantity value of the parameter to be controlled matches the control target value, but also on the condition that the values ​​of other dimensions that make up the state are within a predetermined range.

[0054] As an example of such settings, for example, a requirement for the final target state may be added that the rate of change of the controlled variable of the controlled parameter along the time axis is 0 or near 0. In this way, even if the controlled variable matches the control target value, if it intersects with a gradient (i.e., if it subsequently departs from the control target value), it does not fall into the final target state. Instead, it falls into the final target state only when the controlled variable of the controlled parameter asymptotically matches the control target value or remains near the control target value. As a result, the fluctuation of the controlled variable achieved by control while calculating the state transition tends to remain near the control target value, thereby improving the stability of control.

[0055] Similarly, for other parameters that make up the process state, if the ranges of values ​​that are likely to appear when process control is stable or the ranges of values ​​that should be met from the perspective of process control stability are known, then including these value ranges as additional requirements for the final target state will reflect the known knowledge of process control stability and the required requirements in the control process that is achieved by following state transitions, thereby enabling further multifaceted improvement in control stability, performance, and safety.

[0056] Furthermore, the determination of step St4 may be based on whether or not at least the controlled variable value among the parameters constituting the process state for the transition destination candidate calculated in step St3 is within the interval defined by the current measurement value and the control target value at that time. This type of determination can be expected to be consistent even when the division width in the discretization of the state is coarse, because the size of the region of the final target state does not change depending on the division width but is a single, invariant point, the control target value.

[0057] In addition, in this case, if the final target state has numerical range requirements specified as described above not only for the controlled variables but also for the parameters that constitute other states, the requirement for determining whether the current state is between the final target state may include that for at least one of these requirements, the value of the parameter of the requirement at the transition destination is between the boundary of the range that satisfies the requirement (here, this refers to the boundary of both the upper and lower limits that is closer to the current value) and the current value.

[0058] When the number of judgment requirements increases in this way, unless the added requirements are carefully selected and set, it becomes difficult for the judgment result to become true, and it becomes difficult to efficiently and effectively find the next target state. However, if appropriate signals are selected as the added requirements and the range of values ​​that satisfy the requirements is set appropriately, it is possible to effectively identify the next target state to achieve the control goal, and high controllability can be obtained.

[0059] In step St5, the next target state calculation unit 21a resets the transition destination candidates calculated in step St3 to the pre-transition process state S. Thereafter, step St3 is called again, and the next target state calculation unit 21a recalculates the transition destination candidates based on the pre-transition process state S reset in step St5. This recalculation is repeated until the determination in step St4 becomes true, so that the transition destination candidates are updated every time the determination in step St4 becomes false.

[0060] On the other hand, in step St6, the next target state calculation unit 21a sets the transition destination candidate when the result of step St4 is determined to be true as the target state to be aimed at in the next control cycle.

[0061] In step St7, the command value calculation unit 21b calculates a manipulated variable command value for the controlled element 11 to realize the target state set in step St6, and transmits it to the controlled element 11.

[0062] In step St8, the manipulated variable calculation unit 21 completes a series of processes, including calculation of the next target state and calculation of the manipulated variable command value for realizing the next target state. Note that, since the process in Fig. 2 is performed for each control cycle, after step St8, step St1 of the next control cycle is performed.

[0063] <Effects of the process control method of this embodiment> Next, a mechanism for improving the performance and stability of the process system 1 to follow the control target value by the process control method of this embodiment described with reference to FIG. 2 will be described.

[0064] Figure 3 is a schematic diagram showing the time-series trend of the controlled variable of a certain controlled process. As explained in the "Problem to be Solved by the Invention" section, in process control, there are often cases where the response measured at the measurement end of the controlled variable to be controlled in response to a control operation at the control end is delayed, a process with a large disturbance exists between the measurement end of the controlled variable to be controlled and the control end that executes the control operation, or control of the target process interferes with control of another process.

[0065] In process control, where these issues become apparent, time-series changes in the controlled variable of a controlled object parameter often repeat long-period ups and downs around a set control target value v0 (line L0 in the figure), as exemplified by line L11 in the figure. When operation data or simulation data of a process with such oscillation characteristics is used as training data to train a state transition probability model M on the transition of the process state, the controlled variable value of the next target state (the value of the controlled variable of the controlled object among the parameters constituting the next target state) calculated using the state transition probability model M may also reflect the ups and downs of the training data. In other words, the controlled variable value of the next target state may be farther from the control target value than the current controlled variable value.

[0066] Referring to FIG. 3, first, the controlled variable value v1 at the current time t1 is represented by point B1 on the graph, and this controlled variable value v1 exceeds the control target value v0. If this is set as the pre-transition process state S (step St2) and transition destination candidates are calculated using the state transition probability model M (step St3), the controlled variable value (point B2) corresponding to the transition destination candidate at time t2 may exceed the current controlled variable value v1 (false in step St4). In this case, if point B2 is set as the target state for the next control cycle, the target state will be set even further away from the control target value v0 than the current state, which may cause a decrease in the tracking performance and stability of the process control.

[0067] Therefore, in this embodiment, instead of immediately setting the transition destination candidate calculated using the state transition probability model M as the next target state, the transition destination candidate that satisfies the requirements of step St4 (i.e., a transition destination candidate that is smaller than the current control amount value v1 but does not go below the control target value v0) is set as the next target state (proceed to step St6 only if step St4 is true).

[0068] More specifically, during the control cycle at time t1, a control loop consisting of processes St3, St4, and St5 in FIG. 2 is repeated until a transition destination candidate (point B7) that satisfies the requirements of process St4 is calculated, thereby calculating transition destination candidates (points B2 to B7) for times t2 to t7. During this process, points B2 to B6 (shown as black circles) that do not satisfy the requirements of process St4 and point B7 (shown as a white circle) that satisfies the requirements of process St4 are calculated. However, in process St6, point B7 that satisfies the requirements is set as the next target state, rather than points B2 to B6 that do not. Then, in process St7, a manipulated variable command value that realizes the state of point B7, which is on the way to the final target state (control target value), is calculated, and the manipulated variable 11 is controlled. This allows the controlled variable to converge to the control target value v0 more quickly than if manipulated variable command values ​​that realize the states of points B2 to B6 were sequentially generated and the manipulated variable 11 was sequentially controlled. This improves the tracking performance and stability of process control.

[0069] As described above, by repeatedly calculating the state transition destination and determining whether the transition destination candidate is between the current state and the control target setting state, this embodiment can prevent a decrease in control performance caused by setting the control amount of the next target state to be further away from the control target value v0 than the current control amount value. As a result, stable and highly responsive control performance can be obtained even for processes that have the three problems described above.

[0070] In addition, although FIG. 3 illustrates points B1 to B7 and times t1 to t7, these are generalized calculation results of the state transition destination according to the passage of a certain time, and do not necessarily correspond to the time intervals for each control period or each state transition period.

[0071] As described above, according to this embodiment, stable control with high tracking ability to the control target value can be achieved for processes that are difficult to control, such as when there is a large delay until a change in the controlled variable to be controlled appears in response to the control operation of the process, when there is a large disturbance occurring between the control manipulation end and the controlled variable measurement end, or when control of the target process interferes with control of other processes in the plant. [Example]

[0072] Next, a second embodiment of the present invention will be described with reference to Fig. 4. Note that a duplicated description of points common to the first embodiment will be omitted.

[0073] In the first embodiment, the manipulated variable command value of the process system 1 is calculated in accordance with the flowchart of FIG. 2, but in this embodiment, the manipulated variable command value of the process system 1 is calculated in accordance with the flowchart of FIG.

[0074] The difference between the configurations of the two flowcharts is that the repetitive calculation loop in the first embodiment is formed by steps St3, S4, and S5, whereas the repetitive calculation loop in the present embodiment is formed by steps St3, St3A, St3B, St4, and St5. Step St3A, which is added in the present embodiment, is a step for determining whether the number of repetitive calculations has reached a predetermined lower limit, and step St3B is a step for determining whether the number of repetitive calculations has reached a predetermined upper limit; the details of these steps will be described later. Note that, although not shown in FIG. 4, a counting step is also provided for cumulatively acquiring the number of times step St3 has been executed each time the repetitive calculation is executed.

[0075] If the determination in step St3A is false, that is, while the number of repeated calculations has not reached a predetermined lower limit, regardless of the value of the transition destination candidate calculated in step St3, proceed to step St5, set the transition destination candidate as the pre-transition state for the next calculation, and recalculate the transition destination candidate in step St3.

[0076] On the other hand, if the determination in step St3A becomes true, that is, if the number of times the repeated calculation has been performed reaches a predetermined lower limit, the process proceeds to the next step St3B.

[0077] If the determination in step St3B is false, that is, while the number of times the repeated calculation is performed does not reach the predetermined upper limit, the process proceeds to step St4, and, as in Example 1, it is determined whether the transition destination candidate is between the current state and the final target state.

[0078] On the other hand, if the judgment of step St3B becomes true, that is, if the number of repeated calculations reaches a predetermined upper limit, the process proceeds to step St6, and the transition destination candidate calculated in the most recent step St3 is set as the target state for the next control cycle.

[0079] The reasons for adding steps St3A and St3B and the effects thereof will be described below.

[0080] First, as mentioned above as a problem, in process control, after a control operation is executed by transmitting a manipulated variable command value to the manipulated variable 11 of the process system 1, there is often a delay before the effect of the control operation appears in the measured value of the signal (controlled variable) to be controlled at the measurement terminal 12. In this case, if there is a process between the manipulated variable 11 and the controlled variable measurement terminal 12 that is prone to large disturbances, it becomes even more difficult to identify the delay before the effect of the control operation appears, as mentioned above.

[0081] Under such conditions, for example, if the time required to reach the next target state calculated using the state transition probability model M is short and shorter than the length of time from the operation of the control element to the appearance of an effect on the controlled variable, there is a risk that the effect of the control operation will be exceeded by the influence of process fluctuations, resulting in a situation where control execution cannot keep up with the process fluctuations. In such cases, it is desirable to estimate, from past performance, state transitions for a time longer than the time from the operation of the control element to the appearance of an effect on the controlled variable (delay time), and to calculate a transition destination that is estimated to be desirable for a sufficiently long time ahead.

[0082] By providing a step of presetting and compensating for the lower limit number of iterative calculations of the state transition, as in step St3A, it becomes possible to estimate a desirable transition destination assuming a time period ahead corresponding to this lower limit number of state transitions. This makes it possible to determine the target destination of the state transition while effectively estimating a desirable future state, even for a process in which the appearance of changes in the controlled variable in response to a control operation is prone to be delayed, and to achieve stable control by suppressing degradation of control performance due to delayed response to a control operation.

[0083] Furthermore, in the repeated calculation of the state transition destination, the greater the number of iterations, the longer the time period in the future the state transition is estimated to be. If the number of iterations of the state transition calculation becomes too large, not only does the accuracy of the estimation decrease, but the estimated state may also deviate significantly from the current state.

[0084] Therefore, by setting and limiting the upper limit of the number of repeated calculations of the state transition in advance, as in step St3B, it is possible to prevent a decrease in control performance due to such a decrease in estimation accuracy, and to prevent the control from becoming unstable due to the calculation of a target that will transition to a state that is extremely different from the current state.

[0085] Note that this embodiment has been described as including both process St3A for determining whether the number of repetitions has reached the lower limit and process St3B for determining whether the number of repetitions has reached the upper limit. However, as described above, process St3A for determining the lower limit number of repetitions and the control flow switching based thereon, and process St3B for determining the upper limit number of repetitions and the control flow switching based thereon each have different significances and effects, and therefore, it is also possible to include either process St3A or process St3B depending on the characteristics of the process.

[0086] <How to set the minimum and maximum number of iterations> Next, a method for setting the "lower limit number of times" used in process St3A and the "upper limit number of times" used in process St3B in Fig. 4 will be described with reference to Fig. 5. A display device and an input device are connected to the process control device 2 of this embodiment, which are used by the user when setting the "lower limit number of times" and the "upper limit number of times." The display device is a liquid crystal display or the like, and the input device is a keyboard and mouse, or a touch screen function provided on the display device, or the like.

[0087] 5 shows an example of a setting screen 4 displayed on a display device. As shown here, the setting screen 4 is provided with a setting value input field 41 for inputting the lower limit number of times for process St3A, and a setting value input field 42 for inputting the upper limit number of times for process St3B. The user inputs desired values ​​into these input fields via an input device. Note that in the figure, "minimum number of state transitions" is an item name indicating that this is an input field for the lower limit number of repeated calculations, and "maximum number of state transitions" is an item name indicating that this is an input field for the upper limit number of repeated calculations.

[0088] After entering these values, when the user clicks (or touches, if it is a touch screen) the registration button 43 provided on the setting screen 4, the entered setting values ​​are imported into the process control device 2 as the "lower limit number of times" and "upper limit number of times" to be used in process St3A and process St3B, and are saved in the setting file.

[0089] Thereafter, when the user clicks (or touches in the case of a touch screen) the back button 44 provided on the setting screen 4, the screen transitions to the operation monitoring screen that is normally displayed while the process control system 100 is operating.

[0090] In this embodiment, by providing a setting screen 4 as shown in Figure 5, it is possible to easily set or change the upper and lower limits of the number of repeated calculations of transition destination candidates according to the characteristics of the process system 1.

[0091] For example, when setting a lower limit, the user must determine the relationship between the number of iterations of the transition destination candidate and the time it takes for the control variable to be affected after the control operation, as well as the magnitude of these effects. Therefore, the user must set the lower limit based on the results of this determination, or change the set limit to a more appropriate value if it is inappropriate. On the other hand, when setting an upper limit, an excessive number of iterations of the transition destination candidate may result in side effects such as a decrease in the accuracy of estimating the desired transition destination candidate or a process state that deviates too much from the current process state being calculated as the next target state. Therefore, it is necessary to set an appropriate upper limit to prevent these side effects, or change the set limit to a more appropriate value if it is inappropriate.

[0092] Setting the upper and lower limits on the number of calculations for these transition destination candidates can have a significant impact on the stability and performance of control, especially when the three issues mentioned above for the controlled process are significantly manifesting as control problems. Therefore, making these settings and adjustments visible and easy to perform is effective in improving the operational efficiency of the plant and ensuring that it is always running efficiently.

[0093] Furthermore, when the operating conditions change over time due to long-term operation, or when the process specifications change due to maintenance or partial modification of process equipment, which changes the response characteristics of the process and the requirements for control, the control settings can be fine-tuned in response to these changes, making it easier to continue stable operation and efficient operation.

[0094] Although the setting screen 4 in FIG. 5 displays both setting value input fields 41 and 42, it is not necessary to provide both of these. If either step St3A or step St3B is omitted depending on the characteristics of the process, the corresponding input field may be omitted. [Explanation of symbols]

[0095] 100 Process Control Systems 1 Process System 11 Control end 12 Measurement end 2. Process control equipment 21 Manipulated amount calculation section 21a Next target state calculation unit 21b Command value calculation unit 22 Learning Model 4. Settings screen 41, 42 Setting value input field 43 Registration button 44 Back button

Claims

1. A process control device that controls a process system having an operating end and a measuring end, and calculates an operation amount command value of the operating end based on a measurement value of the measuring end, a state transition probability model that quantifies the probability that a process state will transition to another process state after a predetermined time period; a next target state calculation unit that calculates a target state to be aimed at in the next control period based on the state transition probability model; a command value calculation unit that calculates the manipulated variable command value that achieves the target state and supplies the manipulated variable command value to the manipulated variable, The next target state calculation unit inputting the current process state into the state transition probability model to calculate transition destination candidates; determining whether the transition destination candidate is between the current process state and the final goal process state; If the determination result is false, the transition destination candidate is input into the state transition probability model to recalculate the transition destination candidate; If the determination result is true, the process control device sets the transition destination candidate to the target state.

2. 2. The process control device according to claim 1, the current process state is a process state defined by the value of any one parameter of the manipulated variable command value or the measured value in a current control cycle, or a process state defined by a combination of values ​​of any two or more parameters of the manipulated variable command value or the measured value.

3. 2. The process control device according to claim 1, The process control device according to claim 1, wherein the state transition probability model is calculated based on a weighted average of a plurality of state transition probability models.

4. 2. The process control device according to claim 1, the next target state calculation unit determines whether the recalculation has been repeated a predetermined number of times before determining whether the transition destination candidate is between the current process state and a final target process state, and if the determination result is false, inputs the current transition destination candidate into the state transition probability model and recalculates the transition destination candidate.

5. 5. The process control device according to claim 4, The process control device according to claim 1, wherein the lower limit number of times is a lower limit number of times set by a user in a setting value input field on a setting screen.

6. 2. The process control device according to claim 1, the next target state calculation unit determines whether the recalculation has been repeated a predetermined upper limit number of times before determining whether the transition destination candidate is between the current process state and a final target process state, and sets the current transition destination candidate to the target state if the determination result is true.

7. 7. The process control device according to claim 6, The process control device is characterized in that the upper limit number of times is an upper limit number of times set by a user by inputting an input value into a setting value input field on a setting screen.

8. 1. A process control method for controlling a process system having an operating end and a measuring end, wherein the method calculates an operating amount command value of the operating end based on a measurement value of the measuring end, the method comprising: a step of inputting a current process state into a state transition probability model that quantifies the probability of a process state transitioning to another process state after a predetermined time period, and calculating transition destination candidates; determining whether the transition destination candidate is between the current process state and a final goal process state; If the determination result is false, inputting the transition destination candidate into the state transition probability model and recalculating the transition destination candidate; If the determination result is true, setting the transition destination candidate as a target state to be aimed for in the next control cycle; calculating the manipulated variable command value that achieves the target state and supplying it to the manipulated variable; A process control method comprising:

Images