Process management system and process control method

Abstract

In plant operation support, the system predicts dynamic optimal control for the controlled process and monitors the process state changes resulting from the predicted optimal control. [Solution] The process management system generates control prediction data relating to the control flow for the controlled process based on a control prediction model that outputs a control flow including the operations and sequence of operations for the process until the target value is reached, in response to input of a target value of a physical quantity in the process, and transmits the control prediction data to a control device that controls the equipment. The process management system then generates process prediction data relating to the future predicted value of the physical quantity when the controlled process is controlled based on the control prediction data, based on the process prediction model that outputs a future predicted value of the physical quantity in the process in response to input of the control prediction data and the control prediction data, and outputs the process prediction data from an output device.

Description

Technical Field

[0001] The present invention relates to a process management system and a process management method.

Background Art

[0002] Conventionally, plants such as petrochemical plants, power generation plants, and water treatment plants have been controlled by a DCS (Distributed Control System) in which equipment constituting the plant and a control device for controlling the equipment are connected. Further, the DCS collects and records performance data related to automatic control of the plant and manual control of the plant by intervention of workers.

[0003] In order to be used for plant control, the relationship between an operation performed on the plant and a change in the plant state may be modeled based on past performance data and the like. As a related prior art, there is the technique disclosed in Patent Document 1.

[0004] Also, in order to predict the future state of the plant, a model for predicting future physical quantities in the plant may be constructed. As a related prior art, there is the technique disclosed in Patent Document 2.

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0006] For example, to identify the optimal control of a process to be controlled, one might consider using a model that describes the relationship between operations performed on the plant and changes in the plant's state. However, plant processes can change due to various factors such as adjustments to operating conditions and environmental changes, so it is desirable to be able to predict and monitor how the process changes under the identified control.

[0007] Another approach is to predict and monitor a plant using models that forecast future physical quantities of the controlled process. However, process changes in a plant can be complex, and such models alone may not be able to efficiently or accurately identify the controls needed to reach the target state.

[0008] Therefore, in order to improve plant operation, a system is needed that can efficiently identify the optimal control to achieve the target state in the controlled process, while simultaneously predicting and monitoring the changes in the process state caused by that optimal control.

[0009] The aforementioned problems are not considered in Patent Documents 1 and 2.

[0010] This invention has been made in view of the above points, and aims to predict dynamic optimal control for a controlled process in plant operation support, and to monitor the changes in the process state due to the predicted optimal control. [Means for solving the problem]

[0011] To achieve the above objective, the present invention, in one aspect, is a process management system for managing a process executed in equipment constituting a plant, wherein the process management system comprises a processor and a memory, and the processor generates control prediction data relating to the control flow for the controlled process based on a control prediction model that outputs a control flow including operations on the process and the order of operations until the target value is reached in response to an input of a target value of a physical quantity in the process, transmits the control prediction data to a control device that controls the equipment, and generates process prediction data relating to the future predicted value of the physical quantity when the controlled process is controlled based on the control prediction data, based on a process prediction model that outputs a future predicted value of a physical quantity in the process in response to an input of the control prediction data and the control prediction data, and outputs the process prediction data from an output device. [Effects of the Invention]

[0012] According to the present invention, for example, in plant operation support, it is possible to predict dynamic optimal control for the process to be controlled and to monitor the changes in the process state due to the predicted optimal control. [Brief explanation of the drawing]

[0013] [Figure 1] A diagram showing the configuration of the process management system according to the embodiment. [Figure 2] A diagram showing operational performance data according to the embodiment. [Figure 3] A diagram showing characteristic information according to the embodiment. [Figure 4] A diagram showing a state transition model according to an embodiment. [Figure 5] A figure showing a control prediction model according to an embodiment. [Figure 6] A figure showing control prediction data according to the embodiment. [Figure 7] A diagram showing the process prediction model and characteristic parameters according to the embodiment. [Figure 8] A diagram illustrating the prediction process according to the embodiment. [Figure 9] Figure showing Use Case 1 (Real-time processing) of the prediction processing according to the embodiment. [Figure 10] Figure showing the GUI (Real-time processing) according to Use Case 1 of the embodiment. [Figure 11] Figure showing the GUI (Real-time processing: Comparison with the initial setting flow) according to Use Case 1 of the embodiment. [Figure 12] Figure showing Use Case 2 (Simulation processing) of the embodiment. [Figure 13] Figure showing the GUI (Simulation processing) according to Use Case 2 of the embodiment. [Figure 14] Figure showing Use Case 3 (Automatic control) of the embodiment. [Figure 15] Figure showing the control monitoring processing according to the embodiment. [Figure 16] Figure showing Use Case 4 (Model update) of the embodiment.

Mode for Carrying Out the Invention

[0014] Hereinafter, embodiments of the present invention will be described with reference to the drawings. The embodiments are examples for explaining the present invention, and for the sake of clarity of explanation, appropriate omissions and simplifications have been made. The present invention can be implemented in various other forms. Unless otherwise particularly limited, each component may be singular or plural. In the drawings, the positions, sizes, shapes, ranges, etc. of each component shown are not necessarily representative of the actual positions, sizes, shapes, ranges, etc. in order to facilitate understanding of the invention. Therefore, the present invention is not necessarily limited to the positions, sizes, shapes, ranges, etc. disclosed in the drawings.

[0015] Examples of various types of information may be described using terms such as "table," "list," and "queue," but these types of information may also be represented by other data structures. For example, various types of information such as "XX table," "XX list," and "XX queue" may be referred to as "XX information." When describing identification information, terms such as "identification information," "identifier," "name," "ID," and "number" are used, but these are interchangeable.

[0016] When there are multiple components with the same or similar function, they may be described using the same symbol but with different subscripts. Furthermore, when it is not necessary to distinguish between these multiple components, the subscripts may be omitted in the description.

[0017] In embodiments, processing performed by executing a program may be described. Here, the computer executes the program using a processor (e.g., CPU, GPU) and performs processing defined by the program using memory resources (e.g., memory) and interface devices (e.g., communication ports). Therefore, the main entity performing the processing by executing the program may be the processor. Similarly, the main entity performing the processing by executing the program may be a controller, device, system, computer, or node having a processor. The main entity performing the processing by executing the program may be an arithmetic unit, and may include dedicated circuits that perform specific processing. Here, dedicated circuits include, for example, FPGAs (Field Programmable Gate Arrays), ASICs (Application Specific Integrated Circuits), CPLDs (Complex Programmable Logic Devices), etc.

[0018] The program may be installed on the computer from the program source. The program source may be, for example, a program distribution server or a storage medium readable by the computer. If the program source is a program distribution server, the program distribution server includes a processor and storage resources for storing the program to be distributed, and the processor of the program distribution server may distribute the program to other computers. In addition, in some embodiments, two or more programs may be implemented as a single program, or one program may be implemented as two or more programs.

[0019] [Embodiment] Figure 1 shows the configuration of the process management system 1 according to an embodiment. The plant whose process the process management system 1 controls is assumed to be a transient plant in which the target values ​​of physical quantities measured by the equipment constituting the plant change over time. However, it is not limited to transient plants; it may also be a plant in which the target values ​​of physical quantities do not change over time. Furthermore, the process to be controlled is assumed to be a batch process, but it may also be a continuous process.

[0020] The process management system 1 comprises a memory 11, an input device 12, an output device 13, a processor 14, a storage unit 15, and a communication interface 16. The process management system 1 is connected to one or more pieces of equipment 3 for each process constituting the plant via a control device 2.

[0021] The control device 2 acquires operational performance data 15a of the equipment 3, which includes information on the measured values ​​of physical quantities of the processes performed in the equipment 3, and transmits it to the process management system 1. The control device 2 also transmits control instructions for the equipment 3 received from the process management system 1 to the equipment 3.

[0022] The memory 11 of the process management system 1 is a storage device having a volatile memory area. The input device 12 is a device for inputting administrator operations to the process management system 1, such as a keyboard or mouse. The output device 13 is a device for outputting information from the process management system 1 to the administrator, such as a display or speaker. The processor 14 executes programs in cooperation with the memory 11. The storage unit 15 is a storage unit having a non-volatile memory area. The communication interface 16 is a communication device for the process management system 1 to communicate with the control unit 2.

[0023] The processor 14 includes a data acquisition unit 14a, a control instruction unit 14b, a control prediction unit 14c, a process prediction unit 14d, a model evaluation unit 14e, and a model creation and update unit 14f, which are realized by executing a program.

[0024] The data acquisition unit 14a collects operational performance data 15a, converts it to an appropriate format, and stores it in the storage unit 15. The control instruction unit 14b outputs control commands for each piece of equipment 3 to the control device 2 based on the current control prediction data 15d.

[0025] The control prediction unit 14c uses the control prediction model 15c to perform control predictions based on operating data 15a or virtual conditions. The process prediction unit 14d uses the process prediction model 15e to perform control predictions based on operating data or virtual conditions.

[0026] The model evaluation unit 14e calculates evaluation information 15h regarding the model accuracy of the control prediction model 15c and the process prediction model 15e by comparing the prediction results of the control prediction model 15c and the process prediction model 15e with the actual operating data 15a.

[0027] The model creation and update unit 14f selects training data from the operational performance data 15a, and creates a control prediction model 15c and a process prediction model 15e by learning from the selected training data. The model creation and update unit 14f also selects training data from the operational performance data 15a to achieve the best accuracy under arbitrary conditions, and updates the control prediction model 15c and the process prediction model 15e by learning from the selected training data.

[0028] The memory unit 15 stores operational performance data 15a, characteristic information 15b, control prediction model 15c, control prediction data 15d, process prediction model 15e, process prediction data 15f, characteristic parameters 15g, and evaluation information 15h.

[0029] Control prediction model 15c and process prediction model 15e are provided for each characteristic. Evaluation information 15h is information regarding the accuracy of control prediction model 15c and process prediction model 15e.

[0030] The control prediction model 15c is a model that outputs the manipulated values ​​of equipment 3 to reach target values ​​based on information about target values ​​of physical quantities in the plant process and information about measured values ​​of physical quantities related to the process. The control prediction model 15c is a model that has learned the relationship between information about the manipulated values ​​of equipment 3 included in past operating results and information about measured values ​​of the operating results. The control prediction model 15c has learned multiple state transition models 15c0 until the measured value of the operating result reaches the target value of the physical quantity.

[0031] The control prediction model 15c can select the optimal operation flow (control flow) from among multiple state transition models 15c0 until the measured value of the operation result reaches the target value of the physical quantity. In any case, if any constraint conditions (such as reaching the target value faster, the measured value not exceeding a certain value, or reducing the manipulated amount) are satisfied, the operation flow in question is defined as the "optimal operation flow".

[0032] The process prediction model 15e includes two use cases. Use case 1 is a case where the future is predicted based on the current state. Use case 2 is a case where the future is predicted based on a hypothetical state (simulation mode, or training data augmentation mode for the control prediction model 15c).

[0033] In Use Case 1, the input values ​​of the process prediction model 15e are measured physical quantities (PV value, current value) + operation flow, and the output values ​​are predicted future physical quantities. Specifically, in Use Case 1, future predicted values ​​of process physical quantities are output based on information on measured physical quantities related to the process and information on the operation quantities of equipment 3.

[0034] The process prediction model 15e includes a physical model. The physical model enables the output of predicted values ​​even with limited training data. Furthermore, the process prediction model 15e includes a machine learning model. The machine learning model enables the output of predicted values ​​based on data, even in cases where physical principles cannot be established or where disturbances occur.

[0035] In Use Case 2, the input values ​​of the process prediction model 15e are virtual data of physical quantities (PV values) + operation flow, and the output values ​​are predicted values ​​of future physical quantities.

[0036] (Driving performance data 15a according to this embodiment) Figure 2 shows the operational performance data 15a according to the embodiment. The operational performance data 15a has columns for "Operation No.", "Type", "Equipment", "Date and Time", "Operation", and "Measured Value".

[0037] "Operation No." is the identification information for the operation performed by the plant operator. "Product Type" is the product type (e.g., manufactured goods) handled by the equipment 3 being operated on. "Equipment" is the identification information for the equipment 3 being operated on. "Date and Time" is the date and time when the operation was performed. "Operation" indicates the content of the operator's operation. "Measured Value" is the value of a physical quantity such as flow rate, pressure, or temperature measured by the equipment 3 in question. "Measured Value" includes the values ​​of one or more types of physical quantities. By classifying the "Measured Value" into multiple discrete "states," one of the multiple "states" is associated with the "Operation" in each record of the operation performance data 15a.

[0038] (Characteristic information 15b according to the embodiment) Figure 3 shows characteristic information 15b according to an embodiment. Characteristic information 15b stores the characteristics of the controlled object (e.g., product type, equipment, etc.) and the characteristic types included in each characteristic (a, b, A, B, etc.) in association with each characteristic. In Figure 3, for example, a, b, etc. are defined as characteristic types for product type, and A, B, etc. are defined as characteristic types for equipment 3.

[0039] (State transition model 15c0 according to the embodiment) Figure 4 shows a state transition model 15c0 according to the embodiment. The state transition model 15c0 is the basic model used when creating the control prediction model 15c. The state transition model 15c0 is created for each product type and piece of equipment 3.

[0040] In the example in Figure 4, when operation a1 is performed on the target equipment 3 in state s190, there is a 20% probability of transitioning to state s211, a 70% probability of transitioning to state s212, and a 10% probability of transitioning to state s213 after a predetermined transition time (e.g., 10 seconds later). Furthermore, even in the same state s190, when operation a2 is performed, there is a 55% probability of transitioning to state s233 and a 45% probability of transitioning to state s234. This state transition model 15a1 can be created by discretizing past operating performance data 15a into "states" and then statistically processing the relationship between the operation (set value of a physical quantity) and the state after 10 seconds, which is the transition time.

[0041] (Control and prediction model 15c according to the embodiment) Figure 5 shows the control prediction model 15c according to the embodiment.

[0042] The control prediction model 15c is a model that represents the optimal action (operation) to take in order to reach a target state from a given state in the fastest possible time. The control prediction model 15c can be obtained from the state transition model 15c0 using reinforcement learning or dynamic programming. When using reinforcement learning, a large reward is given when performing an operation from a given state makes it easier to approach the target state.

[0043] The control prediction model 15c stores information about the second state to which the controlled object transitions when a first operation is performed on the controlled object in the first state. For example, in the example shown in the first row of Figure 5, it indicates that in state s190, performing operation a2 is the optimal operation, and as a result, the object transitions to state s233.

[0044] In the state transition model 15c0 in Figure 4, multiple operations that have been performed in the past and the transition probability when each operation is performed are represented for a given state. In contrast, in the control prediction model 15c in Figure 5, only the optimal operation for ultimately transitioning to the target state is represented from among the multiple operations that have been performed in the past.

[0045] As described above, in the learning phase of the control prediction model 15c, a state transition model 15c0 is created from past operating performance data 15a, and the control prediction model 15c is created from the state transition model 15c0. The control prediction model 15c may also store the information necessary to transition from the first state to the second state for each combination of the first and second states.

[0046] The learning method for the control prediction model 15c is as follows. First, the processor 14 acquires operational data 15a related to the operational performance of the equipment 3, including measured values ​​of physical quantities in the process. Next, the processor 14 determines the state of the process from the measured values ​​of physical quantities. Next, the processor 14 classifies the determined state of the process into multiple categories using a predetermined classification process (such as clustering). Next, based on the state of the process classified into multiple categories, the processor 14 generates a state transition model 15c0 that includes one or more states of the process to which it will transition when a certain operation is performed on a process in a certain state, and the probability of transitioning to this state. Next, based on the state transition model 15c0, the processor 14 generates the control prediction model 15c by generating the optimal transition route for the process state to transition from the first state to the second state.

[0047] (Control prediction data 15d according to the embodiment) Figure 6 shows the control prediction data 15d according to the embodiment. Figure 6 shows the control prediction data 15d when transitioning from the initial state s190 to the final target state s355. Using the control prediction model 15c shown in Figure 5, one or more actions, i.e., operations, required when transitioning from the initial state s190 to the final target state s355 are uniquely identified, as shown in Figure 6. The control prediction data 15d includes the operations on the process and the sequence of operations until the target value is reached, in response to the input of a target value for a physical quantity in the process.

[0048] (Process prediction model 15e and characteristic parameters 15g according to the embodiment) Figure 7 shows the characteristic parameters 15g and the process prediction model 15e according to the embodiment. Figure 7 shows an example of the process prediction model 15e and the relationship between the process prediction model 15e and the characteristic parameters 15g.

[0049] The process prediction model 15e is represented by a model constructed using a predetermined theoretical formula. In Figure 7, the predetermined theoretical formula is y=ax 2 The expression is given as +bx+c. In this example, the initial values ​​of the parameters a, b, and c of the process prediction model 15e are given as 1, 0, and 0, respectively.

[0050] Furthermore, it is shown that the parameter correction amounts for process prediction model 15e were calculated as a:0.2, b:0, c:0 for product I, and a:0, b:-0.1, c:0 for equipment A. The parameter correction amounts (correction parameters) for process prediction model 15e are stored in characteristic parameters 15g for each product and each piece of equipment 3.

[0051] Furthermore, the initial values ​​of parameters a, b, and c of the process prediction model 15e were corrected by the respective correction amounts for each product type and each piece of equipment 3. As a result, the corrected parameters a, b, and c of the process prediction model were calculated to be 1.2, -0.1, and 0, respectively.

[0052] Figure 7 illustrates a table in which characteristic parameters 15g are stored for each product type and piece of equipment 3. This table contains characteristic parameters 15g for each characteristic. Furthermore, in Figure 7, the process prediction model 15e is y=ax 2 Assuming the model is represented by +bx+c, the explanation assumed that the parameters consist of three components: a, b, and c. However, as mentioned above, if the process prediction model 15e is represented in other ways, a number of characteristic parameters corresponding to each representation are stored.

[0053] The learning method for the process prediction model 15e is as follows. First, the processor 14 acquires operational performance data 15a for each characteristic of the equipment 3, which includes measured values ​​of physical quantities in the process. Next, the processor 14 acquires characteristics from the operational performance data 15a. Next, for each characteristic, the processor 14 calculates correction parameters to correct the process prediction model 15e so that the actual values ​​of the physical quantities of the process executed by the equipment 3 and the estimated values ​​of the physical quantities estimated by the process prediction model 15e satisfy predetermined conditions. The predetermined conditions are, for example, that the difference between the measured quantity and the estimated quantity of the process prediction model is minimized. Next, the processor 14 corrects the process prediction model 15e using the correction parameters for each characteristic. The process prediction model 15e corrected in this way becomes the created process prediction model 15e.

[0054] (Prediction process according to the embodiment) Figure 8 shows the prediction process according to the embodiment.

[0055] First, in step S11, the data acquisition unit 14a acquires the operating performance data 15a corresponding to a predetermined period from the operating performance data 15a stored in the storage unit 15 for each controlled process.

[0056] Next, in step S12, the control prediction unit 14c obtains the optimal control flow in the control prediction model 15c toward the target value of the controlled process, based on the operational performance data 15a for a predetermined period acquired in step S11 and the control prediction model 15c. Specifically, for the controlled process, the control prediction unit 14c takes the operational performance data 15a corresponding to the predetermined period acquired in step S11 as measured values ​​of physical quantities and inputs them into the control prediction model 15c along with future target values ​​of physical quantities. The control prediction unit 14c then obtains the optimal control flow (control value, control set value, SV value) for the controlled process output from the control prediction model 15c. The control flow is data that includes operations on the process and the order of operations, and arranges the control value, control set value, SV value, etc. in chronological order in the order of execution.

[0057] Next, in step S13, the control prediction unit 14c generates control prediction data 15d (referred to as the first control data) for the process to be controlled based on the control flow acquired in step S12, and stores it in the storage unit 15.

[0058] Next, in step S14, the data acquisition unit 14a acquires the first control data of the controlled process stored in the storage unit 15 in step S13, and the operation performance data 15a corresponding to a predetermined period of the controlled process.

[0059] Next, in step S15, the process prediction unit 14d generates process state change prediction information (process prediction data 15f (referred to as first prediction data)) which includes predictions of changes in physical quantities when the process to be controlled is controlled based on the first control data. The process prediction unit 14d generates the first prediction data based on the first control data acquired in step S14, the operating performance data 15a corresponding to a predetermined period of the process to be controlled, and the process prediction model 15e. The process prediction unit 14d then stores the generated first prediction data in the storage unit 15.

[0060] Next, in step S16, the model evaluation unit 14e acquires the first control data generated in step S13 and the first prediction data generated in step S15, and outputs them from the output device 13.

[0061] (Use case 1 of the prediction processing according to the embodiment (real-time processing)) Figure 9 shows Use Case 1 (Real-time Processing) of the prediction processing according to the embodiment. Use Case 1 (Real-time Processing) is one of the use cases of the prediction processing shown in Figure 8. In Use Case 1 (Real-time Processing), guidance regarding plant control is displayed in real time, for example, while the plant is in operation, and instructions for correcting the plant control are input and reflected in the plant control.

[0062] The control device 2 outputs control instructions to the equipment 3 (S101) and collects various physical quantities (PV values) from the equipment 3 (S102). The data collection unit 14a stores the information collected by the control device 2 as operating performance data 15a in the storage unit 15 (S103).

[0063] The control prediction unit 14c outputs control prediction data 15d based on predetermined information from the operation performance data 15a stored in the storage unit 15 and the control prediction model 15c (S104). The process prediction unit 14d outputs process prediction data 15f based on predetermined information from the operation performance data 15a stored in the storage unit 15, the control prediction data 15d and the process prediction model 15e (S105).

[0064] The control prediction data 15d output by the control prediction unit 14c and the process prediction data 15f output by the process prediction unit 14d are displayed on the administrator screen 17 output from the output device 13 (S106). The plant administrator refers to the display on the administrator screen 17 and inputs correction instructions regarding the control of the equipment 3 from the input device 12 to the control device 2 (S107). The control device 2 controls the equipment 3 based on the corrected control prediction data 15d, which has been modified based on the information corrected by the administrator (S108).

[0065] (GUI13D1 relating to Use Case 1 of the embodiment) Figure 10 shows a GUI (Graphical User Interface) 13D1 (real-time processing) related to Use Case 1 of the embodiment. GUI 13D1 includes a selection field 131, an optimal control screen 132, and a state prediction screen 133. Selection field 131 includes a product type selection field 131a, a lot selection field 131b, and an equipment selection field 131c. A product type is selected in product type selection field 131a. A product lot number is selected in lot selection field 131b. Equipment 3 is selected in equipment selection field 131c. When the product type, product lot number, and equipment 3 are selected and the execute button 131d is pressed, information related to optimal control and predicted state is associated and displayed on the optimal control screen 132 and the state prediction screen 133. When the modify button 131e is pressed, modifications to the product type, product lot number, and equipment 3 become available.

[0066] The optimal control screen 132 shows the actual values ​​of past optimal control flows for the selected product type, product lot number, and raw material input amount for equipment 3 as a solid line, and the predicted future optimal control flow based on the control prediction model 15c as a dashed line. In the future optimal control flow, the raw material input amount changes from 20 to 15 Nm on 4 / 5 at 12:30. 3 / h, 4 / 5 13:30 15→10Nm 3 It has been reduced to / h. And at 15:10 on 4 / 5, the operation of equipment 3 will end.

[0067] The status prediction screen 133 shows the past status with a solid line and the future status predicted based on the process prediction model 15e with a dashed line, for the selected product type, product lot number, and product conversion rate and temperature related to equipment 3. In the future status, the product conversion rate reaches the termination condition at 4 / 5 15:10, and the operation of equipment 3 ends.

[0068] Furthermore, multiple combinations of the optimal control flow (control prediction data) shown by a dashed line on the optimal control screen 132 and the future prediction values ​​(process prediction data) shown by a dashed line on the state prediction screen 133 may be generated and displayed as candidates. The administrator can then input the optimal control flow that they have determined to be the best from among the multiple candidates as feedback input to the equipment 3 via the control device 2.

[0069] (GUI13D1 relating to Use Case 1 of the embodiment) Figure 11 shows GUI13D1 (Real-time processing: Comparison with initial setting flow) related to use case 1 of the embodiment. Compared with Figure 10, Figure 11 further shows the control flow when the process is controlled according to the initial setting flow on the optimal control screen 132 with a dashed line, and further shows the changes in product conversion rate and temperature when the process is controlled according to the initial setting flow on the state prediction screen 133 with a dashed line.

[0070] Referring to Figure 11, it can be seen that implementing the optimal control flow predicted based on the control prediction model 15c instead of the initial setting flow increases the product conversion rate, quickly reaches the termination condition, and maintains a higher temperature, thus demonstrating the significance of implementing the optimal control flow.

[0071] (Use case 2 of the embodiment (simulation processing)) Figure 12 shows Use Case 2 (Simulation Processing) of the embodiment. Use Case 2 (Simulation Processing) is one of the use cases of the prediction processing shown in Figure 8. In Use Case 2 (Simulation Processing), guidance on the simulation of plant control under virtual conditions is displayed, and instructions to modify the virtual conditions are entered and reflected in the simulation.

[0072] Use Case 2 differs from Use Case 1 in that, while the administrator refers to the display on the administrator screen 17, the input device 12 receives a modification to the virtual conditions assumed by the simulation, rather than a modification instruction regarding the control of the equipment 3 (S107'). The modification to the virtual conditions is reflected in the control prediction processing by the control prediction unit 14c.

[0073] (GUI13D1 relating to Use Case 2 of the embodiment) Figure 13 shows the GUI 13D1 (simulation) related to use case 2 of the embodiment. Compared with Figure 10, Figure 13 further shows the control flow when the process is controlled according to the simulation results on the optimal control screen 132 using a dashed line. Also, compared with Figure 10, Figure 13 further shows the changes in product conversion rate and temperature when the process is controlled according to the simulation results on the state prediction screen 133 using a dashed line.

[0074] (Use case 3 of the embodiment (automatic control)) Figure 14 shows Use Case 3 (Automatic Control) of the embodiment. Use Case 3 (Automatic Control) is one of the use cases of the predictive processing shown in Figure 8. In Use Case 3 (Automatic Control), the control prediction data 15d is fed back to the control device 2 and the plant is automatically controlled. At the same time, guidance regarding the control of the plant is displayed in real time, for example, and based on this display, instructions for correcting the control of the plant are input (S107) and reflected in the control of the plant.

[0075] In use case 3, unlike use case 1, correction instructions regarding the control of equipment 3 are input from the input device 12 while referring to the display on the administrator screen 17 (S107), and the control prediction data 15d is fed back to the control device 2, and the plant is automatically controlled (S109).

[0076] (Control and monitoring process according to the embodiment) Figure 15 shows the control monitoring process according to the embodiment. Step S21 of the control monitoring process (Figure 10) corresponds to step S11 of the prediction process (Figure 9). Also, step S22 of the control monitoring process (Figure 10) corresponds to steps S12 to S15 of the prediction process (Figure 9).

[0077] First, in step S21, the model creation and update unit 14f selects training data from the operating performance data 15a for each characteristic. The selection criteria for the training data here are, for example, that the training data has a timestamp for a recent period, but are not limited to this.

[0078] Next, in step S22, the model creation and update unit 14f learns from the training data selected in step S21 and creates a control prediction model 15c and a process prediction model 15e for each characteristic. Then, in step S23, the control prediction unit 14c and the process prediction unit 14d start operating the control prediction model 15c and process prediction model 15e for each characteristic created in step S22.

[0079] The next steps, S24-S27, are model accuracy monitoring processes.

[0080] First, in step S24, the data acquisition unit 14a collects operational performance data 15a for each characteristic and stores it in the storage unit 15. Next, in step S25, the control prediction unit 14c and the process prediction unit 14d use the control prediction model 15c and the process prediction model 15e (operational model) that were put into operation in step S23, and make predictions using the current operational performance data 15a collected in step S24. Then, the control prediction unit 14c and the process prediction unit 14d generate control prediction data 15d and process prediction data 15f.

[0081] Next, in step S26, the model evaluation unit 14e compares the predicted values ​​(control prediction data 15d and process prediction data 15f) generated in step S25 with the actual values ​​based on the current operating performance data 15a collected in step S24.

[0082] Next, in step S27, the model evaluation unit 14e determines whether the error between the predicted value and the actual value compared in step S26 is greater than or equal to a threshold. Specifically, if at least one of the error between the control prediction data 15d and its actual value, and the error between the process prediction data 15f and its actual value is greater than or equal to a threshold (step S27YES), the model evaluation unit 14e proceeds to step S28. On the other hand, if both the error between the control prediction data 15d and its actual value, and the error between the process prediction data 15f and its actual value are less than a threshold (step S27NO), the model evaluation unit 14e returns to step S24.

[0083] The next steps, S28-S31, are the model recreation process.

[0084] First, in step S28, the model creation and update unit 14f selects training data for model reconstruction from the operational performance data 15a for each characteristic. Here, compared with the training data selection conditions in step S21, training data is selected that allows for the reconstruction of the control prediction model and process prediction model, which can make the most accurate predictions, for example, by using more up-to-date data or matching the characteristic information more closely.

[0085] Next, in step S29, the model creation and update unit 14f learns from the training data selected in step S28 and recreates the control prediction model 15c and / or process prediction model 15e for each characteristic. In step S29, the model creation and update unit 14f recreates the control prediction model 15c and process prediction model 15e that were determined in step S27 to have an error between the predicted value and the actual value that is greater than or equal to a threshold.

[0086] Next, in step S30, the model evaluation unit 14e compares the model accuracy of the recreated model created in step S29 with the model currently in operation. Specifically, it compares the model accuracy, such as the error between predicted values ​​and actual values, for the control prediction model 15c and / or process prediction model 15e recreated in step S29.

[0087] Next, in step S31, the model evaluation unit 14e determines whether the recreated model has higher model accuracy than the currently operating model. If the recreated model has higher model accuracy than the currently operating model (step S31 YES), the model evaluation unit 14e moves the process to step S32. On the other hand, if the recreated model has lower model accuracy than the currently operating model (step S31 NO), the model evaluation unit 14e moves the process to step S28.

[0088] In step S28, which is a transition from step S31, the model creation and update unit 14f selects training data for model recreation from the operating performance data 15a for each characteristic. Here, the selection criteria are stricter compared to the selection criteria for training data in the previous execution in step S21, for example, regarding timestamps and characteristics.

[0089] In step S32, the model creation and update unit 14f updates the operational model with the recreated model that was determined to have higher model accuracy than the operational model in step S31.

[0090] (Embodiment Use Case 4 (Model Update)) Figure 16 shows Use Case 4 (Model Update) of the embodiment. Use Case 4 (Model Update) is one of the use cases of the control monitoring process shown in Figure 15. In Use Case 4 (Model Update), the control prediction data 15d and / or process prediction data 15f are recreated and updated according to the monitoring results of the model accuracy of the control prediction data 15d and process prediction data 15f.

[0091] The model evaluation unit 14e monitors the model accuracy of the control prediction data 15d and the process prediction data 15f (S110). The model evaluation unit 14e displays information regarding predicted and actual values ​​based on the control prediction data 15d and the process prediction data 15f (S111) on the administrator screen 17 output from the output device 13.

[0092] The model creation and update unit 14f recreates the target model (S112) if there is a discrepancy between the predicted values ​​based on the control prediction data 15d and / or process prediction data 15f and the measured values ​​(S112). At this time, it selects training data from the operational performance data 15a to improve the model accuracy of the control prediction data 15d and / or process prediction data 15f to be recreated. The control prediction data 15d and / or process prediction data 15f are updated by the recreated model.

[0093] The model creation and update unit 14f may also augment the training data (operational performance data 15a) for the control prediction model 15c and the process prediction model 15e with process prediction data 15f. This augmentation allows for the creation of improved models even when sufficient data has not been accumulated to update the control prediction model 15c and the process prediction model 15e, such as during the initial stages of plant operation or when operating conditions are changed.

[0094] (Effects of the embodiment) In the above-described embodiment, the process management system 1, which manages the processes executed in the equipment 3 constituting the plant, has a processor 14 and a memory 11. The processor 14 generates control prediction data 15d relating to the control flow for the process to be controlled, based on a control prediction model 15c that outputs a control flow in response to input target values ​​of physical quantities in the process. Here, the control flow includes the operations performed on the process and the sequence of operations until the target value is reached.

[0095] Furthermore, based on the process prediction model 15e and the control prediction data 15d, process prediction data 15f is generated, which relates to the future predicted values ​​of physical quantities when the process to be controlled is controlled based on the control prediction data 15d. Here, the process prediction model 15e outputs future predicted values ​​of physical quantities in the process in response to the input of the control prediction data 15d. Then, the information relating the control prediction data 15d and the process prediction data is output from the output device 13.

[0096] Therefore, according to this embodiment, in plant operation support, it is possible to predict dynamic optimal control for the process to be controlled and to monitor the changes in the process state due to the predicted optimal control.

[0097] Furthermore, in the above-described embodiment, the target value of the physical quantity in the process changes over time. Therefore, according to the embodiment, a control flow based on more appropriate control prediction data 15d can be fed back to the equipment 3 in accordance with the target value of the physical quantity that changes over time.

[0098] Furthermore, in the above-described embodiment, the process is a batch process. Therefore, according to the embodiment, a control flow based on more appropriate control prediction data 15d can be fed back to the equipment 3 in accordance with the batch process.

[0099] Furthermore, in the above-described embodiment, control prediction data 15d is transmitted to the control device 2 that controls the equipment 3, and the control device 2 controls the equipment 3 based on the control prediction data 15d. Therefore, according to this embodiment, the control prediction data 15d can be fed back to the equipment 3.

[0100] Furthermore, in the above embodiment, multiple combinations of control prediction data 15d and process prediction data 15f generated in correspondence with the control prediction data 15d are generated as candidates, and the candidates are output from the output device 13. Therefore, according to this embodiment, manual control can be performed in which the administrator manually selects the appropriate control prediction data 15d from the guidance displaying multiple combinations of control prediction data 15d and process prediction data 15f and inputs it as feedback to the equipment 3.

[0101] Furthermore, in the above embodiment, information regarding the administrator's modification of the control prediction data 15d output from the output device 13 is received, and process prediction data 15f is generated representing the process to be controlled when the process is controlled based on the modified control prediction data 15d according to the modification information. Then, the generated process prediction data 15f is output from the output device 13. Thus, according to this embodiment, automatic control can be performed by automatically modifying the control prediction data 15d and automatically feeding it back to the equipment 3.

[0102] Furthermore, in the above-described embodiment, operational performance data 15a relating to the actual operation of the equipment 3 when controlled based on the control prediction data 15d is acquired. Then, based on the process prediction data 15f and the operational performance data 15a, it is determined whether or not the control prediction model 15c and the process prediction model 15e need to be updated. Thus, according to this embodiment, the model accuracy of the control prediction model 15c and the process prediction model 15e can be monitored and model degradation can be detected.

[0103] Furthermore, in the above-described embodiment, the recreated control prediction model, which is recreated to update the control prediction model 15c, has higher model accuracy than the control prediction model 15c, and the training data is selected to achieve the highest accuracy among the training data selection patterns. Then, the recreated control prediction model is created by training on the selected training data, and the created recreated control prediction model is used to update the control prediction model 15c. The process prediction model 15e is the same as the control prediction model 15c. Therefore, according to this embodiment, when model degradation of the control prediction model 15c and the process prediction model 15e is detected, the decrease in their model accuracy can be suppressed.

[0104] Furthermore, in the above-described embodiment, when creating the recreated control prediction model and the recreated process prediction model, in addition to the training data selected for recreation, process prediction data 15f predicted based on the control prediction model 15c and process prediction model 15e before recreation is also trained. Thus, according to this embodiment, the training data (operational performance data 15a) can be augmented with process prediction data 15f.

[0105] The present invention is not limited to the embodiments described above, and in the implementation stage, the components can be modified and implemented without departing from the gist of the invention, or the multiple components disclosed in the embodiments can be appropriately combined. [Explanation of Symbols]

[0106] 1: Process management system, 2: Control device, 3: Equipment, 11: Memory, 13: Output device, 14: Processor, 14a: Data acquisition unit, 14b: Control instruction unit, 14c: Control prediction unit, 14d: Process prediction unit, 14e: Model evaluation unit, 14f: Model creation and update unit, 15a: Operating performance data, 15a1: State transition model, 15b: Characteristic information, 15c: Control prediction model, 15c0: State transition model, 15d: Control prediction data, 15e: Process prediction model, 15f: Process prediction data, 15g: Characteristic parameters, 15h: Evaluation information.

Claims

1. A process management system that manages processes executed in equipment and machinery that constitute a plant, The aforementioned process management system has a processor and memory, The aforementioned processor, Based on a control prediction model that outputs a control flow including the operations performed on the process and the order of those operations until the target value is reached, in response to an input of a target value for a physical quantity in the process, control prediction data relating to the control flow for the controlled process is generated. The control prediction data is transmitted to the control device that controls the equipment. Based on the process prediction model that outputs predicted future values ​​of physical quantities in the process in response to the input of the control prediction data, process prediction data relating to the predicted future values ​​of physical quantities when the process to be controlled is controlled based on the control prediction data is generated. The aforementioned process prediction data is output from the output device. A process management system characterized by the following features.

2. A process management system according to claim 1, The target value of the aforementioned physical quantity is a value that changes over time. A process management system characterized by the following features.

3. A process management system according to claim 1, The process to be controlled is a batch process. A process management system characterized by the following features.

4. A process management system according to claim 1, The aforementioned processor, The device receives information related to the correction of the control prediction data output from the output device, The process to be controlled is controlled based on the control prediction data modified based on the information related to the modification, and process prediction data is generated. The information relating to the control prediction data and the process prediction data is associated and output from the output device. A process management system characterized by the following features.

5. A process management system according to claim 1, The aforementioned processor, Based on the control prediction data, operational performance data relating to the operational performance of the equipment is acquired. Based on the process prediction data and the actual operating data, determine whether or not the control prediction model needs to be updated. A process management system characterized by the following features.

6. A process management system according to claim 5, The aforementioned processor, The training data for the recreated control prediction model, which is recreated to update the aforementioned control prediction model, is selected from the operational performance data such that the recreated control prediction model has higher accuracy than the original control prediction model and achieves the highest accuracy among the training data selection patterns. The selected training data is used to create the re-creation control prediction model. The control prediction model is updated using the recreated control prediction model that was created. A process management system characterized by the following features.

7. A process management system according to claim 6, The aforementioned processor, When creating the aforementioned recreation control prediction model, the process prediction data is used in addition to the selected training data. A process management system characterized by the following features.

8. A process management system according to claim 1, The aforementioned processor, Based on the control prediction data, operational performance data relating to the operational performance of the equipment is acquired. Based on the process prediction data and the actual operating data, determine whether or not the process prediction model needs to be updated. A process management system characterized by the following features.

9. A process management system according to claim 8, The aforementioned processor, The recreated process prediction model, which is recreated to update the aforementioned process prediction model, is selected from the operational performance data such that the recreated process prediction model has higher accuracy than the original process prediction model and achieves the highest accuracy among the selection patterns for training data. The selected training data is used to create the reproduction process prediction model. The process prediction model is updated using the recreated process prediction model that was created. A process management system characterized by the following features.

10. A process management system according to claim 9, The aforementioned processor, When creating the aforementioned reproduction process prediction model, the process prediction data is used in addition to the selected training data. A process management system characterized by the following features.

11. A process management system according to claim 1, The aforementioned processor, In the process described above, operational performance data relating to the operational performance of the equipment, including the measured values ​​of the physical quantities, is acquired. The state of the process is determined from the measured values ​​of the aforementioned physical quantities. The determined state of the process is classified into multiple categories by a predetermined classification process. Based on the states of the processes classified into the aforementioned multiple categories, a state transition model is generated that includes one or more states of the process to which it transitions when a certain operation is performed on the process in a certain state, and the transition probability to that state. Based on the state transition model, the control prediction model is generated by generating the optimal transition route in which the state of the process transitions from the first state to the second state. A process management system characterized by the following features.

12. A process management system according to claim 1, The aforementioned processor, In the process described above, operational performance data is obtained for each characteristic of the equipment relating to the operational performance of the equipment, including the measured values ​​of the physical quantities. The characteristics are obtained from the aforementioned operating performance data, For each characteristic included in the operational performance data, correction parameters are calculated to correct the process prediction model such that the actual value of the physical quantity in the process executed by the equipment and the estimated value of the physical quantity estimated by the process prediction model constructed with a predetermined theoretical formula and initial values ​​satisfy predetermined conditions. The process prediction model is created by correcting the process prediction model using the correction parameters for each characteristic. A process management system characterized by the following features.

13. A process management system according to claim 1, The aforementioned processor, We acquire operational performance data related to the operational performance of the aforementioned equipment. A re-creation control prediction model is created by training the training data selected from the aforementioned operational performance data and the aforementioned process prediction data. The control prediction model is updated using the recreated control prediction model that was created. A process management system characterized by the following features.

14. A process management system that manages processes executed in equipment and machinery that constitute a plant, The aforementioned process management system has a processor and memory, The aforementioned processor, Based on a control prediction model that outputs a control flow including the operations performed on the process and the order of those operations until the target value is reached, in response to an input of a target value for a physical quantity in the process, control prediction data relating to the control flow for the controlled process is generated. Based on the process prediction model that outputs predicted future values ​​of physical quantities in the process in response to the input of the control prediction data, process prediction data relating to the predicted future values ​​of physical quantities when the process to be controlled is controlled based on the control prediction data is generated. The information relating to the control prediction data and the process prediction data is associated and output from the output device. We acquire operational performance data related to the operational performance of the aforementioned equipment. A re-creation control prediction model is created by training the training data selected from the aforementioned operational performance data and the aforementioned process prediction data. The control prediction model is updated using the recreated control prediction model that was created. A process management system characterized by the following features.

15. A process management method performed by a process management system that manages processes executed in equipment and machinery constituting a plant, The aforementioned process management system has a processor and memory, The aforementioned processor, Based on a control prediction model that outputs a control flow including the operations performed on the process and the order of those operations until the target value is reached, in response to an input of a target value for a physical quantity in the process, control prediction data relating to the control flow for the controlled process is generated. The control prediction data is transmitted to the control device that controls the equipment. Based on the process prediction model that outputs predicted future values ​​of physical quantities in the process in response to the input of the control prediction data, process prediction data relating to the predicted future values ​​of physical quantities when the process to be controlled is controlled based on the control prediction data is generated. The aforementioned process prediction data is output from the output device. A process management method characterized by having each of the following processes.

16. A process management method performed by a process management system that manages processes executed in equipment and machinery constituting a plant, The aforementioned process management system has a processor and memory, The aforementioned processor, Based on a control prediction model that outputs a control flow including the operations performed on the process and the order of those operations until the target value is reached, in response to an input of a target value for a physical quantity in the process, control prediction data relating to the control flow for the controlled process is generated. Based on the process prediction model that outputs predicted future values ​​of physical quantities in the process in response to the input of the control prediction data, process prediction data relating to the predicted future values ​​of physical quantities when the process to be controlled is controlled based on the control prediction data is generated. The control prediction data and the process prediction data are output from the output device. We acquire operational performance data related to the operational performance of the aforementioned equipment. A re-creation control prediction model is created by training the training data selected from the aforementioned operational performance data and the aforementioned process prediction data. The control prediction model is updated using the recreated control prediction model that was created. A process management method characterized by having each of the following processes.

Images