Simulator system and simulator method

Abstract

To easily perform a simulation of a plant.SOLUTION: A simulator system includes: a plant model processing section 131 that generates an apparatus model, which is a model of an apparatus, on the basis of a performance calculation formula for calculating performance of apparatuses constituting a plant, and generating a plant model that is a model of the plant by connecting the apparatus model; a reinforcement learning section 132 that estimates a control constant that is a constant used for the performance calculation formula by reinforcement learning on the basis of the plant model; and an operation simulator system 2 that performs a simulation of a plant using the plant model in which the estimated control constant is set.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a simulator system and a simulator method. [Background technology]

[0002] As of June 2021, the Ministry of Economy, Trade and Industry's Coal-Fired Power Plant Working Group is considering policy responses to achieve the "phase-out of inefficient coal by 2030."

[0003] In the 5th Basic Energy Plan, coal is positioned as a key baseload power source in Japan, with the issue of large greenhouse gas emissions, but the geopolitical risk of coal being the lowest among fossil fuels.In addition, coal has the lowest unit price per heat unit among fossil fuels, and is therefore currently evaluated as an important baseload power source with excellent stability and economic value.

[0004] When setting efficiency targets for coal-fired power plants, the definition of inefficient coal-fired power plants was clearly stated as SC (Super Critical) or lower (high efficiency is USC (Ultra Super Critical) or higher), with a line drawn based on power generation method. Meanwhile, the current benchmark system for thermal power generation efficiency under the "Act on the Rational Use of Energy" (Energy Conservation Act) uses actual power generation efficiency as an indicator, rather than power generation method classification.

[0005] The benchmark target for thermal power generation efficiency is a target for all thermal power generation facilities (coal, LNG (Liquefied Natural Gas), and oil). Therefore, operators who own all coal, LNG, and oil thermal power plants can achieve the benchmark if thermal power generation using fuels other than coal is highly efficient. Therefore, the effectiveness of achieving the target using only coal-fired power plants is not guaranteed, and it is necessary to set new indicators and targets for coal-fired power plants alone.

[0006] Patent Document 1 discloses an anomaly detection device, simulator, plant monitoring system, an anomaly detection method, and a program, which "includes an anomaly detection device that includes an acquisition unit that acquires actual machine measurement values ​​measured from the actual machine of the plant and acquires control commands from a control device, a simulation unit that calculates a simulator calculation value indicating the state of the actual machine of the plant based on the control commands, an anomaly judgment unit that judges that an anomaly has occurred in the actual machine when the degree of deviation of the actual machine measurement value from the calculated simulator calculation value exceeds a predetermined judgment threshold, and a notification processing unit that notifies the occurrence of the anomaly when it is judged that the anomaly has occurred" (see abstract).

[0007] Patent Document 2 discloses a power plant performance management device that is characterized by comprising: "plant input means for scanning and inputting process data of a power plant; plant setting determination means for determining a specific setting state of the plant based on the input process data; plant data collection means for collecting the process data when the determination result of the plant setting determination means is the setting state; data storage means for saving the collected data in a large-capacity storage device; search means for extracting data from the large-capacity storage device; calculation means for aggregating the searched data and calculating plant performance; and output means for outputting the current calculation result and a transition from the past calculation results" (see claim 1).

[0008] Patent Document 3 discloses a power system control device, a power system control system, a power system control method, and a power system control program, as follows: "The data collection server 200 collects and stores data on various electrical quantities in each of a plurality of subsystems included in the power system at predetermined intervals. The digital twin generation unit 102 generates a simulation model of the power system including a model reproducing each subsystem based on the data of various electrical quantities stored by the data collection server 200, and updates the model based on changes in the various electrical quantities. The simulation execution unit 102 executes a simulation based on the data and simulation model collected by the data collection server 200. The result notification unit 103 notifies the results of the simulation executed by the simulation execution unit 102" (see abstract). [Prior art documents] [Patent documents]

[0009] [Patent Document 1] JP 2020-129158 A [Patent Document 2] Special Publication No. 08-016618 [Patent Document 3] JP 2019-154201 A Summary of the Invention [Problem to be solved by the invention]

[0010] The technology described in Patent Document 1 does not take into consideration plant performance diagnosis, and the technology described in Patent Document 2 does not include any description of analysis of factors.

[0011] The technology described in Patent Document 3 does not take into account plant performance diagnosis and digital twin aspects in thermal and hydroelectric power plants and renewable energy power generation. In addition, the system analysis simulation described in Patent Document 3 reproduces the system state at the time of a system accident, and is not a simulation for calculating plant performance such as boiler efficiency, turbine efficiency, and heat balance.

[0012] The present invention has been made in view of the above background, and an object of the present invention is to easily perform a plant simulation. [Means for solving the problem]

[0013] In order to solve the above-mentioned problems, the present invention provides a plant model generation unit that generates an equipment model that is a model of equipment constituting a plant based on a performance calculation formula for calculating the performance of the equipment, and generates a plant model that is a model of the plant by connecting the equipment models; a control constant estimation unit that estimates control constants that are constants used in the performance calculation formula by reinforcement learning based on the plant model; and a simulator unit that performs a simulation of the plant using the plant model in which the estimated control constants are set. a hierarchical structure processing unit that extracts a hierarchical structure of calculation symbols used in the performance calculation formula based on the performance calculation formula and displays the hierarchical structure on a display unit; The present invention is characterized by having the following. Other solutions will be described in the embodiments as appropriate. Effect of the Invention

[0014] According to the present invention, a plant simulation can be easily performed. [Brief description of the drawings]

[0015] [Figure 1] 1 is a diagram illustrating an example of the configuration of a plant simulator system according to an embodiment of the present invention. [Diagram 2] FIG. 2 is a diagram showing an overall process performed in the plant simulator system according to the present embodiment. [Diagram 3] FIG. 11 is a diagram showing a detailed procedure of statistical processing. [Figure 4] 13 is a flowchart showing a procedure for an analysis support process. [Diagram 5] FIG. 13 is a diagram showing decomposition of a calculation formula specification, a function specification, and an input point specification. [Figure 6] 10 is a flowchart showing the procedure of a hierarchical structure search process performed in this embodiment. [Figure 7] FIG. 13 is a diagram showing an example of a calculation symbol. [Figure 8] FIG. 11 is a diagram showing the procedure of an analysis support display process performed in this embodiment. [Figure 9] FIG. 13 is a diagram showing an example of a formula analysis screen displayed in the analysis support display process. [Figure 10] 1 is a flowchart showing a procedure for generating a plant model. [Figure 11] FIG. 11 is a diagram showing a procedure of a simple device model generation process. [Figure 12] FIG. 1 is a diagram showing a valve model as an example of a simple device model. [Figure 13] FIG. 13 is a diagram illustrating an example of an input / output list. [Figure 14] 4 is an example of a plant model generated by a plant model generation process. [Figure 15] FIG. 10 is a diagram relating to a confirmation process of a plant model. [Figure 16A] FIG. 4 is a diagram showing a procedure of a control constant estimation process performed in the present embodiment. [Figure 16B] 11A and 11B are diagrams illustrating a specific example of an operation parameter determination process. [Figure 17] FIG. 11 is a diagram illustrating an example of control parameters. [Figure 18] FIG. 11 is a diagram showing an example of an operation simulator system using a plant model generated as a result of the present embodiment. [Figure 19] FIG. 1 is a diagram illustrating an example of a driving simulator system according to a comparative example. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0016] Next, a mode for carrying out the present invention (referred to as an "embodiment") will be described in detail with reference to the drawings as appropriate.

[0017] A plant simulator system Z (see FIG. 1: simulator system) in this embodiment generates a plant model 520 (see FIG. 14 and FIG. 16A), which is a model of a plant, by connecting a simplified equipment model 350 (see FIG. 11), which is an equipment model generated based on a performance calculation formula. The equipment is equipment that constitutes a plant. Then, the plant simulator system Z uses this plant model 520 to estimate a control constant, which is a constant in the performance calculation formula, by reinforcement learning. Thereafter, the estimated control constant is applied to the plant model 520, thereby providing an operation simulator system 2 (see FIG. 1) that simulates the plant.

[0018] In detail, in this embodiment, an operation simulation of an existing thermal power plant, a hydroelectric power plant, and a renewable energy power plant is performed, and it becomes possible to judge the plant performance under a predetermined condition (see FIG. 18). In addition, in this embodiment, a hierarchical structure of calculation symbols in a performance calculation formula is searched, and an operation level, which is the difference between a current value and a target value (best value), is calculated, and it is possible to support the analysis of factors such as efficiency reduction in a plant (see FIG. 9). In addition, in this embodiment, a control constant is estimated based on reinforcement learning from actual data of the plant (operation data 301 (see FIG. 2)) (see FIG. 16A). The calculation symbols are variables used in the performance calculation formula. The control constant is a constant in the calculation formula expressing the simplified equipment model 350. The hierarchical structure and the best value will be described later.

[0019] Although a thermal power plant is assumed as an application target of this embodiment, the application target of this embodiment is not limited to a thermal power plant, and may be other plants. In this embodiment, a thermal power plant is simply referred to as a plant. In this embodiment, an actual plant is also referred to as a plant as appropriate.

[0020] [System and overall processing] First, the configuration and overall processing of a plant simulator system Z according to this embodiment will be described with reference to FIGS.

[0021] (system) First, the configuration of a plant simulator system Z according to this embodiment will be shown with reference to FIG. FIG. 1 is a diagram showing an example of the configuration of a plant simulator system Z according to this embodiment. As shown in FIG. 1, the plant simulator system Z includes a simulator generating device 1 and an operation simulator system (simulator section) 2. The simulator generating device 1 generates a plant model 520 (see FIG. 16A) based on performance data acquired from a plant (hereinafter referred to as operation data 301 (see FIG. 2)) and a preset performance calculation formula for equipment. The equipment is equipment that constitutes the plant. Then, the operation simulator system 2 performs an operation simulation based on the generated plant model 520.

[0022] (Simulator generator 1) The simulator generating device 1 is configured by a PC (Personal Computer) or the like, and includes a memory 100, a CPU (Central Processing Unit) 11, and a storage device 12 including an HDD (Hard Disk Drive), an SDD (Slid State Drive), etc. The simulator generating device 1 further includes an input device 13, a display device (display unit) 14, and a communication device 15.

[0023] A program stored in the storage device 12 is loaded into the memory 100, and the loaded program is executed by the CPU 11, thereby realizing a statistical processing unit 110, an analysis support processing unit 120, a control constant estimation unit 130, an operating parameter determination unit 140, and an operating sequence determination unit 150.

[0024] The statistical processing unit 110 performs an interpolation process and the like on the operation data 301 (see FIG. 2) acquired from the plant. The operation data 301 is actual data (past data) such as flow rate and pressure acquired from an actual plant. The analysis support processing unit 120 includes a hierarchical structure search unit (hierarchical structure processing unit) 121 and an analysis support display processing unit (hierarchical structure processing unit, action level processing unit) 122. The hierarchical structure search unit 121 searches for and outputs the hierarchical structure of the calculation symbols used in the performance calculation formula using the interpolated operating data 301. The hierarchical structure will be described later, but it is the hierarchical structure of the calculation symbols used in the performance calculation formula.

[0025] The analysis support display processing unit 122 displays the hierarchical structure of the calculation symbols searched for by the hierarchical structure search unit 121 on the display device 14. The analysis support display processing unit 122 also calculates an operation level for analyzing the operation efficiency, etc. of the equipment constituting the plant, and displays it on the display device 14. The hierarchical structure and operation level will be described later. Note that hereinafter, the equipment constituting the plant may be simply referred to as equipment.

[0026] The control constant estimation unit 130 includes a plant model processing unit (plant model generation unit) 131 and a reinforcement learning unit (control constant estimation unit) 132 . The plant model processing unit 131 generates a pseudo plant model (plant model 520 (see FIG. 16A)) using the search results by the hierarchical structure search unit 121 and the like. The reinforcement learning unit 132 estimates the control constants of the plant model 520 by reinforcement learning.

[0027] Based on the control constants, the operation parameter determination unit 140 calculates operation parameters (see FIG. 2) of the plant model 520. The operation parameters are, for example, control values ​​of a damper in a thermal power plant. The operation sequence determination unit 150 determines an operation sequence (see FIG. 2) of the plant model 520 based on control constants and operation parameters. The operation sequence is, for example, the order in which the devices constituting the plant operate.

[0028] By applying control constants, operation parameters, operation sequences, etc. to the plant model 520, an operation simulator system 2 (digital twin) for the plant is constructed.

[0029] [Full Processing] Fig. 2 is a diagram showing the overall processing performed in the plant simulator system Z in this embodiment. Fig. 1 will be referred to as appropriate. Fig. 2 shows an overview of the overall processing performed in the plant simulator system Z, and details of each process will be described later. The plant simulator system Z is composed of the following processes. (A1) The statistical processing unit 110 performs statistical processing (S1) that utilizes a statistical model 311 from the driving data 301. "Utilizing a statistical model 311 from the driving data 301" means generating a statistical model 311 for the driving data 301 and using this statistical model 311 to interpolate missing parts of the driving data 301. Specifically, the statistical processing unit 110 uses the optimal statistical model 311 to interpolate the driving data 301.

[0030] (A2) The analysis support processing unit 120 performs analysis support processing (S2) to support the user in analyzing the plant by displaying on the display device 14 (calculation formula analysis screen 400) the hierarchical structure of calculation symbols used in the performance calculation formula stored in the performance calculation formula information 320 and the operation levels for determining the operating states of the equipment that constitutes the plant.

[0031] (A3) The control constant estimation unit 130 performs a control constant estimation process (S3) to estimate a control constant (control constant information 304) that is looked up in reverse from a preset performance formula. "Estimating a control constant that is looked up in reverse from the performance formula 320" means estimating a control constant based on the performance formula. The performance formula is a formula for calculating the performance value of the equipment that constitutes the plant, and is, for example, a formula that indicates power generation efficiency.

[0032] Specifically, a plant model 520 (see FIG. 16A), which is a pseudo plant model, is generated based on the performance calculation formula by the plant model processing unit 131. Then, the control constant estimation unit 130 estimates control constants that enable highly efficient control based on the plant model 520. In this manner, the control constants are estimated based on the performance calculation formula.

[0033] (A4) An operation parameter / operation sequence determination process (S4) is performed to determine operation parameters (operation parameter information 305) and operation sequences (operation sequence information 306) for the control constants estimated by the operation parameter determination unit 140 and the operation sequence determination unit 150.

[0034] (A5) The driving simulator system 2 constructs a driving simulator system 2 (digital twin) based on the determined control constants, operation parameters, and operation sequence. The user adds elements to the constructed driving simulator system 2, taking into consideration progress prediction and support for abnormalities and unexpected events in the driving scenario. Then, the user issues an instruction to the driving simulator system 2, whereby the driving simulator system 2 performs a simulation process, which is an operation simulation of the plant model 520 (S5: simulation step). As a result, the driving simulator system 2 performs a simulation of the plant using the plant model 520 in which the estimated control constants are set. (A6) The user evaluates whether the result of the current driving simulation is good or not based on the result from the driving simulator system 2 (S6).

[0035] If the result of step S6 is optimal, the actual equipment is operated based on the plant model 520. If not optimal, the processes of steps S1 to S6 are repeated, thereby achieving highly efficient plant settings.

[0036] Below, an overview will be given of each of steps S1 to S6 in Fig. 2. A more detailed description of steps S1 to S6 will be given later.

[0037] In statistical processing (S1) that utilizes a statistical model 311 from driving data 301 of (A1), the statistical processing unit 110 generates pseudo driving data 302 (see FIG. 3) using a commonly used base radius calculation method or the like. The pseudo driving data 302 is data for complementing missing parts of the driving data 301. In the base radius calculation method, it is possible to generate a statistical model 311 that corresponds to the variability of the driving data 301 by using a calculation method such as the CV method, the radius formula, the proposed method, or the improved proposed method. The statistical processing unit 110 generates the pseudo driving data 302 using the optimal statistical model 311 from among the generated statistical models 311.

[0038] In the analysis support process (S2) of (A2), the hierarchical structure search unit 121 constituting the analysis support processing unit 120 searches for a hierarchical structure in the performance calculation formula (performance calculation formula information 320) for carrying out performance calculation. That is, the analysis support processing unit 120 extracts a hierarchical structure of calculation symbols used in the performance calculation formula based on the performance calculation formula, and displays it on the display device 14.

[0039] In addition, in the analysis support process (S2), the analysis support display processing unit 122 constituting the analysis support processing unit 120 calculates an operation level for determining the operation state of the equipment constituting the plant. The analysis support display processing unit 122 displays the calculated operation level on the display device 14, and visualizes the hierarchical structure searched by the hierarchical structure search unit 121 in a hierarchical diagram showing the hierarchical structure of calculation symbols (calculation formula analysis screen 400) and presents it to the user. This enables visualization in the so-called performance calculation formula structure, and enables the user to analyze factors such as the operation efficiency of the equipment constituting the plant. In addition, the operation level, which is the difference between the best value of the equipment and the operating value (current value or comparison value), is displayed on the display device 14. The best value is calculated based on the performance calculation formula and is the output when the operation efficiency of the equipment is the best. The current value and comparison value will be described later. Then, the user can analyze the displayed contents to analyze factors such as each hierarchical diagram and the decrease in the operation efficiency of the equipment in the performance calculation formula. The operation level is an index for determining the operation state of the plant. That is, the analysis support processing unit 120 calculates an operation level that is an index for determining the operating state of the plant, and displays the calculated operation level on the display device 14 together with a hierarchical structure.

[0040] (A3) In a control constant estimation process (S3) in which a control constant is estimated by performing a reverse lookup from a performance formula, a plant model processing unit 131 constituting the control constant estimation unit 130 generates a pseudo plant model (plant model 520 (see FIG. 16A)) by using the performance formula. Then, a reinforcement learning unit 132 constituting the control constant estimation unit 130 estimates a control constant (control constant information 304) of the plant model 520 by using reinforcement learning.

[0041] Specifically, the plant model processing unit 131 first generates a simplified equipment model 350 (see FIG. 11 ), which is a model of an equipment constituting a plant, based on a performance calculation formula for calculating the performance of the equipment. Next, the plant model processing unit 131 generates a plant model 520, which is a model of the plant, by connecting the simplified equipment models 350. Next, the reinforcement learning unit 132 estimates control constants, which are constants used in the performance calculation formula, by reinforcement learning based on the plant model 520.

[0042] The control constants are estimated by reinforcement learning to achieve highly efficient operation. If multiple control constant candidates are estimated, the reinforcement learning unit 132 may perform a process to determine the control constant that achieves the most efficient operation.

[0043] In the operation parameter / operation sequence determination process (S4) of (A4), first, the operation parameter determination unit 140 determines operation parameters (operation parameter information 305) that realize highly efficient operation of the simulated plant model 520. Furthermore, the operation sequence determination unit 150 determines the estimated control constants and the operation sequence (operation sequence information 306) for the operation parameters.

[0044] (A5) In the simulation process (S5), the estimated control constants, operation parameters, and operation sequence are applied to the plant model 520, thereby constructing an operation simulator system 2 for the plant. Furthermore, the operation data 301 and the pseudo operation data 302 generated by the statistical model 311 are set in the operation simulator system 2, thereby starting an operation simulation of the real plant by the operation simulator system 2. In other words, the operation simulator system 2 performs a simulation of the plant using the plant model 520 in which the estimated control constants are set.

[0045] In the simulation process, the user simulates how to reduce fuel consumption and exhaust gases in the plant. Furthermore, the user can change the control constants and the like in the operation simulator system 2 to try out how the efficiency of the plant changes. Furthermore, in the simulation process, the user can add elements to the training scenario that take into account the progress prediction and countermeasure support for abnormal signs and unexpected events, providing a training solution that realizes an unprecedented high-efficiency plant. In other words, it is possible to try out control values ​​that are difficult to set in an actual plant.

[0046] As described above, the user evaluates whether the result of the current driving simulation is good or not based on the result by the driving simulator system 2 (S6). If the evaluation result shows that the result of the driving simulation is not good, the process returns to step S1. If the result of the simulation process (S5) shows that the result of the driving simulation is good, the settings of the plant model 520 are reflected in the actual plant.

[0047] Below, a detailed description will be given of each process in steps S1 to S6 shown in Fig. 2. In the following description, Fig. 1 will be referred to as appropriate.

[0048] [Details of statistical processing (step S1 in Figure 2)] FIG. 3 is a diagram showing a detailed procedure of the statistical processing shown in step S1 of FIG. As shown in Fig. 3, the statistical processing unit 110 generates a statistical model 311 of the operation data 301 based on the operation data 301 acquired from the plant. Then, the statistical processing unit 110 generates pseudo operation data 302, which is estimated data for complementing missing parts of the operation data 301, based on the statistical model 311. Then, the statistical processing unit 110 can interpolate the operation data 301 (black circles) with the pseudo operation data 302 (x marks) as shown in graph G1 in Fig. 3 by combining the operation data 301 and the pseudo operation data 302. In graph G1, the horizontal axis indicates the operation time, and the vertical axis indicates the process value. The process value is the output value of an instrument in the plant. In graph G1, the dashed vertical lines indicate the boundary (division) of the data to be interpolated.

[0049] The statistical model 311 for generating the simulated driving data 302 may use a method such as a. CV method, b. radial method, c. proposed method, or d. improved proposed method. The procedures for the CV method, the radius method, the proposed method, and the improved proposed method are shown below.

[0050] The calculation method for the a.CV method is as follows. (a-1) The statistical processing unit 110 sets the base radius uniformly for all bases. A base is an item of the operating data 301. For example, it is temperature, the amount of fuel, the amount of air, etc. The base radius is the range of the operating data 301 to be used. For example, it is the range of the amount of fuel to be used. Once the base radius is determined, the base diameter, i.e., the range of the operating data 301 to be used, is determined.

[0051] (a-2) After (a-1) is performed, the statistical processing unit 110 determines the basis by the following algorithm. (a-2-1) The statistical processing unit 110 divides the data samples of the driving data 301 used to construct the statistical model 311 into two groups. (a-2-2) The statistical processing unit 110 generates candidates for the basis radius for each basis. (a-2-3) The statistical processing unit 110 constructs a statistical model 311 using sample data from one group of the divided data samples, and evaluates the accuracy using sample data from the remaining group. (a-3) The statistical processing unit 110 repeats the process of (a-2) to select a base radius that provides a desired accuracy. The desired value is a value that is set in advance.

[0052] b. The calculation method for the radius formula is as follows: (b-1) The statistical processing unit 110 sets the basis radius uniformly for all bases. (b-2) The statistical processing unit 110 determines the basis using the following algorithm. (b-2-1) The statistical processing unit 110 calculates the distance between the input data (operation parameter data) of the operation data 301 used to construct the statistical model 311. The distance between the input data is the distance between the input data in the coordinate space spanned by the input variables. (b-2-2) The statistical processing unit 110 determines the base average value based on the maximum distance calculated in the process of (b-2-1).

[0053] c. The calculation method of the proposed method is as follows. (c-1) The statistical processing unit 110 sets a basis radius for each basis. (c-2) The statistical processing unit 110 defines sparsity information individually for the driving data 301. The sparsity information is the number of data items per unit space in a coordinate space formed by the data. (c-3) The statistical processing unit 110 adjusts the basis radius by changing the neighboring data with respect to the calculated sparseness. The neighboring data is data that is at a predetermined distance from the arbitrary data in the coordinate space formed by the data.

[0054] d. The calculation method of the proposed improved method is as follows. (d-1) The statistical processing unit 110 sets a prescribed radius for each output variable (output data of the statistical model 311) for each basis function, and adjusts the basis radius for each basis radius. The basis function is a function that constitutes the statistical model 311 to be constructed. (d-2) The statistical processing unit 110 adjusts the prescribed radius based on the coverage of the input space and the output variable values ​​of the model construction data. The coverage is the degree to which the basis functions cover the input space.

[0055] The statistical processing unit 110 generates simulated driving data 302 based on the base radius value, base average value, and optimal value of the radius parameter generated by methods a to d. The statistical processing unit 110 can interpolate driving data 301 by using the simulated driving data 302. That is, the statistical processing unit 110 searches for an optimal statistical model 311 based on the base radius and the like obtained by methods a to d, and generates simulated driving data 302 based on the optimal statistical model 311. Then, the statistical processing unit 110 performs interpolation of the driving data 301 using the simulated driving data 302. The generation of the simulated driving data 302 is not limited to the methods a to d, and a method such as linear interpolation may be used.

[0056] [Details of analysis support process (step S2 in Figure 2)] FIG. 4 is a flowchart showing the procedure of the analysis support process shown in step S2 of FIG. As shown in Fig. 4, the analysis support process is composed of a hierarchical structure search process (S21) and an analysis support display process (S22). The analysis support process visualizes the hierarchy of calculation symbols used in the performance calculation formula (see Fig. 2). Details of the hierarchical structure search process in step S21 will be described later, but as shown in Fig. 4, a hierarchical structure search unit 121 constituting the analysis support processing unit 120 searches the hierarchical structure of calculation symbols based on multiple performance calculation formulas (performance calculation formula information 320). The hierarchical structure will be described later. Details of the analysis support information display process in step S22 will be described later, but as shown in FIG. 4, the analysis support processing unit 120 displays information based on the results of the hierarchical structure search process, the operation level, the operating data 301, etc. on the display device 14.

[0057] [Details of hierarchical structure search process (step S21 in FIG. 4)] 5 to 7 are diagrams for explaining the hierarchical structure search process shown in step S21 of FIG. Prior to the hierarchical structure search process, the hierarchical structure search unit 121 decomposes the performance calculation formula in the following procedure. First, as shown in FIG. 5, the performance formula refers to a plurality of formula specifications 321, function specifications 322, and input point specifications 323. The formula specifications 321 store information about the performance formula itself and information about the input points (input variables) used. The function specifications 322 store information about functions. A function is a function included in a performance formula, and the input-output relationship is known in advance (such as a sigmoid function). The input point specifications 323 store information about the input points used in the performance formula. In general, it is difficult for a user to determine the relationship between performance formulas at a glance just by referring to the performance formula. Incidentally, the basis functions used when deriving the statistical model 311 are different from the functions included in the function specifications 322.

[0058] For example, it is assumed that the following formulas (1) and (2) showing power generation efficiency are present as examples of performance calculation formulas. P = (BINOUT / 100)×T (1) PLOSS = (BLOSS / 100)×T (2)

[0059] In equations (1) and (2), "P" is the generating end efficiency based on the input-output method, and "BINOUT" is the boiler efficiency based on the input-output method. Additionally, "T" is the turbine room efficiency, "PLOSS" is the generating end efficiency based on the loss method, and "BLOSS" is the boiler efficiency based on the loss method. Although T is a common input point (input variable) in equations (1) and (2), it is difficult for a computer to infer the relationship between equations (1) and (2) by simply comparing equations (1) and (2).

[0060] Therefore, in this embodiment, the analysis support processing unit 120 decomposes the performance formula to refer to the relationship between each performance formula. Then, using the formula information 331 and definition information 332, which are the results of the decomposition, as keys, the relationship with the input point information 333 is understood, and the relationship between the performance formulas is made clear. The decomposition process of the performance formula is described below.

[0061] (B1) The analysis support processor 120 decomposes the calculation formula specification 321 to obtain calculation formula information 331 including the following information: Calculation symbols, names, calculation sequence, calculation cycle, calculation time In the calculation formula information 331, the calculation symbol and calculation order are search keys for linking with the input point ID described later. In other words, the calculation formula information 331 constitutes part of a relational database with the calculation symbol and calculation order as keys. The search keys in the calculation formula information 331 are written in parentheses in Fig. 5. The same applies to the search keys in the definition information 332, input point information 333, and function information 334 described later.

[0062] The calculation symbol is a variable such as "P", "BINOUT", and so on in formula (1) and formula (2). When searching such a hierarchical structure, the variable is called a calculation symbol. The name is the name of the function indicated by the calculation symbol. For example, the name of the calculation symbol "P" in formula (1) and formula (2) is "generating end efficiency by input / output method". The calculation order is the order in which each performance calculation formula is calculated. For example, formula (1) is calculated followed by formula (2). The calculation cycle is the cycle in which the calculation of the performance calculation formula is performed. The calculation cycle is, for example, calculation is performed at a predetermined time interval. The calculation time is the time required for the calculation of the performance calculation formula. It is preferable that the calculation cycle and the calculation time are stored with information when the calculation of the performance calculation formula is actually performed.

[0063] (B2) The analysis support processing unit 120 obtains the following definition information 332 by decomposing the calculation formula specification 321. · Calculation symbol (definition formula), name, calculation formula, calculation symbol (reference formula) Of the definition information 332, the calculation symbol (definition formula) and the calculation symbol (reference formula) are search keys for linking with an input point ID, which will be described later. The calculation symbol (definition formula) indicates the definition of the calculation symbol. The name is the functional name of the calculation symbol (definition formula). The calculation formula is the performance calculation formula in which the calculation symbol (definition formula) is used. The calculation symbol (reference formula) is the calculation symbol included in the calculation symbol (definition formula). Note that the calculation symbol shown in (B1) corresponds to both the calculation symbol (definition formula) and the calculation symbol (reference formula) in (B2).

[0064] (B3) The analysis support processing unit 120 acquires the following input point information 333 by decomposing the input point specifications 323. Note that the information contained in the input point information 333 shown below is information stored in the input point information 333. Input point ID, name, unit The input point ID is a number that is uniquely assigned to an input point. Incidentally, calculation symbols include input variables and output variables, but the input point is the input variable. "BINOUT", "T", "BLOSS", etc. in formulas (1) and (2) correspond to input points. When searching such a hierarchical structure, the input variable is called the input point. The name is the functional name of the input point. The unit is the unit of the value assigned to the variable indicated by the input point number.

[0065] (B4) The analysis support processing unit 120 obtains the following function information 334 by decomposing the function specification 322. Note that the function information 334 shown below is information included in the function specification 322. Function ID, name, function formula Among the function information 334, the function ID is a search key for linking with the input point ID. As described above, a function indicates the input / output relationship of a given device. A function ID is a number that is uniquely assigned to a function. A function formula is a calculation formula that a function indicates.

[0066] (Hierarchical structure search processing) 6 is a flowchart showing the procedure of the hierarchical structure search process performed in this embodiment. Note that the flowchart shown in FIG. 6 shows detailed processing of step S21 in FIG.

[0067] The hierarchical structure search unit 121 makes a list (including sorting in calculation order) of the search keys in the calculation formula information 331 (S211). Specifically, the hierarchical structure search unit 121 sorts the calculation formula information 331 in calculation order, and then extracts a list of calculation symbols, which are one of the search keys. The list output in step S211 is called a calculation symbol list.

[0068] The hierarchical structure search unit 121 creates a list of the calculation symbols (definition formulas) in the definition information 332 (S212). Specifically, the hierarchical structure search unit 121 obtains and lists the calculation symbols (definition formulas) in the definition information 332 from the definition information 332. The list output in step S212 is referred to as a definition formula list.

[0069] The hierarchical structure search unit 121 adds the calculation symbol (reference formula) of the definition information 332 to the definition formula list (list) (S213). At this time, the hierarchical structure search unit 121 obtains one calculation symbol (reference formula) from the definition information 332, associates it with the calculation symbol (definition formula) stored in step S212, and stores it in the definition formula list. At this time, the calculation symbol (definition formula) and the calculation symbol (reference formula) are associated in accordance with the correspondence with the calculation symbol (definition formula) in the definition information 332.

[0070] Next, the hierarchical structure search unit 121 adds the function ID in the function information 334 to the definition formula list (list) (S214). Specifically, the hierarchical structure search unit 121 obtains one function ID from the function information 334 and adds it to the definition formula list (list).

[0071] Then, the hierarchical structure search unit 121 judges whether or not all of the calculation symbols (reference formulas) in the definition information 332 and the function IDs in the function information 334 have been added to the definition formula list (S215). If all of the calculation symbols (reference formulas) and function IDs have not been added to the list (S215→No), hierarchical structure search unit 121 returns the process to step S213.

[0072] If all of the calculation symbols (reference formulas) and function IDs have been added to the list (S215→Yes), the hierarchical structure search unit 121 checks each of the lists generated in steps S211 to S215 against the input point IDs in the input point information 333, and outputs the formula list information 330 as a final list (S216). In the formula list information 330, the formula list, the definition formula list, and the function list are linked to each other by the input point IDs. That is, in step S216, the hierarchical structure search unit 121 refers to the input point information 333, and adds the input point IDs corresponding to the information (calculation symbols, functions) of the formula list, the definition formula list, and the function list in association with the formula list, the definition formula list, and the function list. As a result, the calculation list, the definition formula list, and the function list are linked to each other by the input point IDs. Furthermore, the calculation formula list information 330 is also linked to calculation formula information 331, definition information 332, input point information 333, and function information 334 via the calculation list, definition formula list, and function list.

[0073] The formula list information 330 thus generated includes information such as the calculation symbols, names, units, formula IDs, and calculation order.

[0074] Through the processing shown in FIG. 5 and FIG. 6, the hierarchical structure search unit 121 extracts the hierarchical structure of the calculation symbols used in the performance calculation formula based on the performance calculation formula.

[0075] Fig. 7 is a diagram showing an example of a calculation symbol. Note that the example shown in Fig. 7 is an example of a calculation symbol, and is different from the calculation formula specification 321, function specification 322, input point specification 323, calculation formula information 331, definition information 332, input point information 333, function information 334 shown in Fig. 5, and each list shown in Fig. 6. In the example shown in FIG. 7, the No., calculation symbol, name, unit, input / output function information (I / O / F), and input point / function ID are shown. The input / output function information stores information on whether the target calculation symbol is used for output, input, or as a function. For example, the calculation symbol "P" is output (OUT) as the generating end efficiency as shown in the above-mentioned formula (1). Also, the calculation symbol "BINOUT" is input (IN) to the generating end efficiency as shown in the formula (1). Note that although there are no examples of calculation symbols used as functions in FIG. 7, when the target calculation symbol is a function, "F" is stored in the input / output function information (I / O / F). Incidentally, the calculation symbols and names in FIG. 7 are the same as those used in formulas (1) and (2).

[0076] The input point / function ID stores the input point ID and the function ID. In the example shown in Figure 7, the calculation symbol itself is used as the input point / function ID. Note that "P" is the output, but when it is used as an input point in another performance calculation formula, it is used as "P".

[0077] (Analysis support display processing) FIG. 8 is a diagram showing the procedure of the analysis support display process (step S22 in FIG. 4) performed in this embodiment. This process uses the calculation formula list information 330 generated in Fig. 6. Calculation formula list information 330 generated in the hierarchical structure calculation process shown in Fig. 6 lists calculation symbols and calculation orders.

[0078] First, the analysis support display processing unit 122 performs an operation level calculation process (S221). In step S221, the analysis support display processing unit 122 acquires the operation data 301, and calculates the operation level for each device by setting the best value of the device and the current value or the comparison value. As described above, the best value is calculated based on the performance calculation formula, and is the output when the operation efficiency of the device is the best. The current value and the comparison value are used for the data of the operation data 301. The current value is the current value of the operation data 301 acquired from the device. The comparison value includes the difference between the operation data 301 at a predetermined time and the current value. In this way, the analysis support display processing unit 122 calculates the operation level such as the difference or deviation between the best value of the device and the operation value (current value or comparison value) acquired from the device. That is, the analysis support display processing unit 122 calculates the operation level based on the best value of the device and the operation value (current value or comparison value) acquired from the device. In this way, the analysis support display processing unit 122 calculates the operation level which is an index for determining the operation state of the plant.

[0079] Next, the analysis support display processing unit 122 performs visualization display processing (S222). The analysis support display processing unit 122 hierarchically displays calculation symbols in the visualization display area 402 shown in Fig. 9 according to the calculation order included in the calculation formula list information 330. Furthermore, in step S222, the analysis support display processing unit 122 displays the action level calculated in step S221, etc. in the action level determination area 401 shown in Fig. 9. Then, the user performs an analysis operation (S223), which will be described later.

[0080] (Calculation formula analysis screen 400) FIG. 9 is a diagram showing an example of a formula analysis screen 400 displayed in the analysis support display process shown in FIG. As shown in FIG. 9, the calculation formula analysis screen 400 has an action level judgment area 401 and a visualization display area 402. The action level is displayed in the action level judgment area 401. In addition to the action level, the action level judgment area 401 may display the current value or comparison value of the operation data 301, the best value, etc. Furthermore, the visualization display area 402 displays the hierarchical structure of the calculation symbols, etc. Here, the hierarchical structure of the calculation symbols will be explained. The hierarchical structure of the calculation symbols indicates the relationship between the calculation symbols. For example, it is assumed that a certain calculation symbol "XA" is calculated by formula A, and the calculation symbol "XB" is used in the formula A. Furthermore, it is assumed that the calculation symbol "XB" is calculated by formula B, which is different from formula (A), and the calculation symbol "XC" is used in the formula B. In other words, it is assumed that the following relationship holds.

[0081] XA = f(XB) XB = f(XC)

[0082] In such a case, a calculation symbol "XB" exists as a subordinate structure of the calculation symbol "XA," and a calculation symbol "XC" exists as a subordinate structure of the calculation symbol "XB." In such a case, the hierarchical relationship of the calculation symbols "XA," "XB," and "XC" is displayed in the format shown in the visualization display area 402 of Fig. 9. Note that the display format in the visualization display area 402 of Fig. 9 is just an example, and the hierarchical structure of the calculation symbols may be displayed in other display formats.

[0083] In this way, the analysis support display processing unit 122 displays the hierarchical structure of the calculation symbols on the display device 14. In addition, the analysis support display processing unit 122 displays the calculated operation level together with the hierarchical structure on the display device 14. By displaying the operation level together with the hierarchical structure, the user can compare the input point ID referred to by the performance calculation formula with its value and perform analysis (S223 in FIG. 8). In the analysis work (S223) in FIG. 8, the user checks the operation status such as the operation efficiency of the equipment based on the operation level, and determines the interrelationship between the equipment and the system based on the hierarchical display. Specifically, the user checks the operation status such as the efficiency of the equipment by referring to the operation level displayed in the operation level judgment area 401. Then, if there is an equipment whose efficiency has decreased, the user analyzes the cause of the efficiency decrease by referring to the hierarchical structure of the calculation symbols displayed in the visualization display area 402.

[0084] Furthermore, in order to allow the user to perform analysis, any performance formula or input point can be excluded. For example, in the case of generating end efficiency, the generator output or integrated value is used. In such a case, it becomes possible to narrow down (reduce) the equipment and systems that make up the plant by eliminating external factors such as operating conditions, seasonal fluctuations, and simple increase values. Note that the exclusion of a performance formula or input point is performed by the user selecting the performance formula or information of the input point displayed on the formula analysis screen 400, and then instructing the user to exclude it.

[0085] [Control constant estimation process (step S3 in Fig. 2) and operation parameter / operation sequence determination process (step S4 in Fig. 2)] The procedure of the control constant estimation process shown in step S3 of FIG. 2 will be described with reference to FIGS.

[0086] First, the control constant estimation process (step S3 in FIG. 2) will be described. In the control constant estimation process, a plant model 520 (see FIG. 16A) is generated prior to the estimation of the control constants. (Plant model generation process) FIG. 10 is a flowchart showing the procedure for generating a plant model 520 (see FIG. 16A) that is generated by combining the simplified device models 350. The plant model processing unit 131 generates a simplified equipment model 350 (see FIG. 11) based on the performance calculation formula (S311: plant model generation step). Note that the simplified equipment model 350 is a model of the equipment constituting the plant, but may also include information on the equipment system. The plant model processing unit 131 sets input and output connections for the simplified device model 350 to be generated, thereby generating a plant model 520. The input and output connections are set based on the calculation formula list information 330 generated in the process of FIG.

[0087] The plant model processing unit 131 generates an input / output list by listing the input / output relationships of the simplified device model 350 (S312: plant model generating step). The input / output list will be described later.

[0088] The plant model processing unit 131 executes calculation processing in the plant model 520 generated in step S312 (S313). Specifically, the operation data 301 and the simulated operation data 302 generated in the processing of FIG. Then, the plant model processing unit 131 outputs the calculation result of the calculation process performed in step S313 to a file (S314), and the user checks the result (S315). In step S315, it is only necessary to determine whether the calculation by the generated plant model 520 is processed normally. Whether the calculation is processed normally means whether an error is output or an abnormal value is output or not. In short, in step S315, the user checks whether the generated plant model 520 is normal or not.

[0089] (Simple device model generation process) FIG. 11 is a diagram showing the procedure of the simple device model generation process carried out in step S311 of FIG. The performance calculation formula is composed of a combination of a plurality of calculation formulas and functions. In this embodiment, this concept is used to construct the simplified equipment model 350 by performing calculations to simulate the components of the plant, such as a pressure unit, a heat exchange unit, a flow rate generation unit, etc.

[0090] For example, the generator simplified device model 350 is generated by using the generating end efficiency shown in formula (1) as a formula (function) representing the generator simplified device model 350.

[0091] The simulated parts (equipment) required to construct the simplified equipment model 350 are as follows. The simulated parts shown below show a part of all the simulated parts. Incidentally, the following functions are functions stored in the calculation formula specification 321 in Fig. 5. The calculation formulas shown below are also used as performance calculation formulas.

[0092] (C1) Machine simulation Pressure vessel calculation formula Pump calculation formula Number of valve calculation formulas ·Temperature calculation formula Enthalpy calculation formula ·Calorie calculation formula ·Flow rate calculation formula (C2) Control simulation PI control formula ·Operation end calculation formula (C3) Analysis simulation Least squares calculation formula ·Linear calculation formula Nonlinear calculation formula (C4) Constant simulation Compatibility table Constant value

[0093] Reference numeral 340 in FIG. 11 indicates a calculation model of the pressure-flow characteristics of the representative pressure section, heat exchange section, and flow generation section as an example of a calculation model constituting the simplified equipment model 350. The representative pressure section is any representative pressure section among the pressure sections. For example, the representative pressure section receives input flow rate "Fi" (kg / s) and input enthalpy "Hi" (kJ / kg). The volume "V" (m 3 ), the output flow rate "Fo" (kg / s) and output enthalpy "Ho" (kJ / kg) are output. In the heat exchange section, the input flow rate "Fi" (kg / s), input enthalpy "Hi" (kJ / kg), and heat input "Qi" (kJ / s) are input. Then, by relating the pressure "P" (kPa) in the heat exchange section to the input, the output flow rate "Fo" (kg / s), output enthalpy "Ho" (kJ / kg), and heat dissipation amount "Qo" (kJ / s) are output. In the flow generation section, the input pressure "Pi" (kPa), the rotation speed (rpm) of the motor, etc., and the valve opening (-) are input. Note that the "-" in the valve opening indicates that the valve is closed. Then, by relating the pressure-flow characteristic: f(x) to the input, the output pressure Po (kPa) is output. Incidentally, the flow rate is calculated from the rotation speed and the valve opening.

[0094] 11, the constructed simple device model 350 is stored as model information 351. The model information 351 is linked to a calculation formula specification 352 and a function specification 353. The calculation formula specification 352 and the function specification 353 linked to the model information 351 have the same content as the calculation formula specification 321 and the function specification 322 shown in Fig. 5. The calculation formula specification 352 and the function specification 353 are generated when the simple device model 350 is constructed.

[0095] In addition, (C2) Control simulation and (C3) Analysis simulation are also assumed to be expressed by performance calculation formulas.

[0096] (An example of the simple equipment model 350) FIG. 12 is a diagram showing a valve model 360 as an example of the simplified device model 350. The valve model 360 shown in FIG. 12 outputs a valve flow rate "F" according to the following equation (11) using the valve inlet pressure "P1", the valve outlet pressure "P2", the valve opening degree "A", and the valve constant "C".

[0097] F = C × (P1-P2) × A (11)

[0098] In addition, "C" in equation (11), which is a calculation formula expressing the valve model 360 (simplified equipment model 350), is a control constant. Moreover, equation (11) is also a performance calculation formula for determining the magnitude of the valve flow rate. Therefore, the control constant ("C" in equation (11)) is a constant used in the performance calculation formula. In this way, the plant model processing unit 131 generates the simplified equipment model (equipment model) 350, which is a model of the equipment, based on the performance calculation formula for calculating the performance of the equipment that constitutes the plant.

[0099] Incidentally, the constant values ​​in the correspondence table shown in (C4) above are constants of the plant model 520 (see FIG. 16A) and are different from the control constants. Also, the value of "C" in equation (11) can be changed (however, it is a constant because it cannot be changed once it is determined), but "100" in equations (1) and (2) cannot be changed and is therefore not a control constant.

[0100] FIG. 13 is a diagram showing an example of the input / output list generated in step S312 of FIG. The input / output list has fields for No., device name, IN / OUT, PID, source PID, and initial value. No. is a number assigned to the connection relationship. The device name is the name of the target device. IN / OUT indicates whether the target information is input or output. Note that for items with numbers at the end, such as "OUT1" and "OUT2" in No. 1 and No. 2 in Fig. 13, the number at the end indicates the number of the output terminal of the device. The same is true for "IN1" and "IN3" in No. 3 and No. 4.

[0101] PID is the input point ID. The source PID is the PID (input point ID) that is the source of the connection for the target device. For example, it shows that the No. 3 device "Z2" is connected to No. 1's "Y1-1". In other words, the output of "Y1-1" is the input of "X1".

[0102] The initial value is the initial value of the target information. The initial value is set by the user. For information for which no initial value is set, the initial value column is left blank.

[0103] As described above, the connections of the simplified device model 350 are set based on the calculation formula list information 330 generated in FIG.

[0104] FIG. 14 shows an example of a plant model 520 generated by the plant model generation process. As shown in Fig. 14, the plant model 520 is configured by connecting a plurality of simplified equipment models 350. For ease of understanding, Fig. 14 illustrates the connection relationships of the equipment indicated by the simplified equipment models 350, but in reality, the connection relationships of each simplified equipment model 350 are shown in the input / output list shown in Fig. 13.

[0105] In this manner, the plant model processing unit 131 connects the simplified device models 350 to generate a plant model 520, which is a model of the plant.

[0106] FIG. 15 is a diagram related to the confirmation process of the plant model 520. 15, in the confirmation process of the plant model 520, the plant model processing unit 131 generates the pseudo-operation data 302 using the statistical model 311 generated from the operation data 301. Note that the statistical model 311 used in the process shown in Fig. 15 may be the statistical model 311 used in the statistical process of Fig. 3 (step S1 of Fig. 2).

[0107] Then, the plant model processing unit 131 applies the operation data 301 and the simulated operation data 302 to the plant model 520 to calculate the operation of the plant model 520 (calculation process: S313). This calculation process corresponds to step S313 in FIG. 10. Then, the user checks the result of the calculation process (S315 in FIG. 10). When the user checks the result of the calculation process, it is sufficient to check whether the plant model 520 generated as described above operates normally.

[0108] (Control constant estimation process) FIG. 16A is a diagram showing the procedure of the control constant estimation process performed in this embodiment. 16A, the control constant estimation unit 130 estimates control constants that will result in highly efficient operation by using the generated plant model 520. Whether or not the estimated control constants result in highly efficient operation is generally determined by performing an analysis using one of the following methods. (D1) A trial calculation of the performance of the statistical model 311 is performed using the simulated driving data 302. (D2) Estimation is performed using a simplified equipment model 350. (D3) A trial calculation is performed using the plant model 520. In all of these techniques, there are fixed values ​​that serve as control constants.

[0109] For example, the performance calculation formula includes a control constant required for performing a trial calculation of the performance of (D1). By changing this control constant, highly efficient operation may be achieved. Similarly, (D2) and (D3) may also be highly efficient operation by changing the control constant in the generated plant model 520. In this embodiment, the control constant that realizes such highly efficient operation is estimated using reinforcement learning.

[0110] Hereinafter, the control constant estimation process performed in this embodiment will be described with reference to FIG. 16A. The calculation process (S313) in Fig. 16A is the calculation process (S313) in Fig. 10 and Fig. 15. That is, it is confirmed that the plant model 520 is normally generated by the calculation process (S313) using the operation data 301 and the simulated operation data 302.

[0111] In this embodiment, reinforcement learning is used to find the control constants. Reinforcement learning is a learning method that continues to search for a method to maximize the value obtained under given conditions through trial and error. Reinforcement learning dynamically changes the constants and control constants of the plant model 520 to search for the control constants that result in the most efficient operation. As described above, the constants of the plant model 520 are the correspondence table and constant values ​​shown in (C4) above.

[0112] Furthermore, the operation parameter determination unit 140 determines the operation parameters by determining the operation conditions of the target device or control terminal based on the control constants that result in the most efficient operation obtained by reinforcement learning. Furthermore, the operation sequence determination unit 150 determines the operation sequence based on the control constants and operation parameters obtained by reinforcement learning. By repeating this cycle, it is possible to optimize the input and output data of reinforcement learning, and determine the state in which highly efficient operation will be achieved.

[0113] Specifically, the reinforcement learning unit 132 of the control constant estimation unit 130 acquires operation parameter setting data 541 and guidance condition data 542. The operation parameter setting data 541 stores information regarding which device is to be used, etc. The guidance condition data 542 stores input values ​​to be input to the devices, operation ratios of each device, etc.

[0114] Next, the reinforcement learning unit 132 performs a reinforcement learning process to execute reinforcement learning (S321: control constant estimation step). Specifically, the reinforcement learning unit 132 inputs the acquired operation parameter setting data 541 and guidance condition data 542 to the plant model 520. Then, the reinforcement learning unit 132 obtains the control constants and the constants of the plant model 520. In this way, the reinforcement learning unit 132 estimates the control constants, which are constants used in the performance calculation formula, by reinforcement learning based on the plant model 520.

[0115] The reinforcement learning unit 132 outputs learning result data 562 and control result data 563 every time reinforcement learning processing is performed. The learning result data 562 stores parameters output as a result of reinforcement learning. The control result data 563 output as a result of reinforcement learning stores control values ​​of each simplified device model 350 in the plant model 520. The control values ​​are, for example, output values ​​of the simplified device models 350 output every time reinforcement learning is performed, and are different from control constants.

[0116] [Operation parameters and operation sequence determination process (step S4 in Figure 2)] The following is an explanation of the operation parameter / operation sequence determination process (step S4 in FIG. 2). Next, the operation parameter determination unit 140 performs an operation parameter determination process (S401). In the operation parameter determination process, a simplified equipment statistical model 550 (see FIG. 16B) is generated based on each constant (constants of the plant model 520, control constants) obtained in the reinforcement learning process.

[0117] Specifically, the operation parameter determination unit 140 applies the constants of the plant model 520, the control constant 304, etc., determined by the reinforcement learning process to the plant model 520 constructed using the condition-specific model input data 521, the condition-specific model output data 522, and the control constant setting data 523 for each simple equipment model 350 output when generating the plant model 520. The condition-specific model input data 521 is input data based on a predetermined condition in the simple equipment model 350. The operation data 301, etc. may be used as the condition-specific model input data 521. The condition-specific model output data 522 is an output of the simple equipment model 350 when the condition-specific model input data 521 is input. The control constant setting data 523 is setting data for narrowing down the control constant to one when there are multiple control constants.

[0118] Then, the operation parameter determination unit 140 generates a simple equipment statistical model 550 which is a statistical model for each simple equipment model 350 in the plant model 520 to which the control constants 304 and the like obtained by the reinforcement learning process are applied. In this way, the simple equipment statistical model 550 is different from the statistical model 311 shown in Fig. 3. The operation parameter determination unit 140 determines the operation parameters (see Fig. 2) by evaluating the simple equipment statistical model 550. The method of determining the operation parameters will be described later.

[0119] Then, as a result of the operation parameter determination process (S401), model characteristic data 561 is output. The model characteristic data 561 stores information on the simplified equipment statistical model 550 shown in Fig. 16B, information on the reference numeral 551, and the like.

[0120] Furthermore, as a result of the reinforcement learning process (S321) and the operation parameter determination process (S401), control parameters (control parameter information 571) are output. The control parameters will be described later, and include information on control constants and operation parameters.

[0121] The operation sequence determination unit 150 performs an operation sequence determination process to determine an operation sequence (see FIG. 2) based on the control parameters (S402).

[0122] Then, the operation sequence determination unit 150 evaluates the plant model 520 using the determined control constants, operation parameters, operation sequence, etc., and if the evaluation is low, reinforcement learning is performed again by the reinforcement learning unit 132 using the plant model 520 (arrow indicated by symbol 581).

[0123] FIG. 16B is a diagram showing a specific example of the operation parameter determination process (S401 in FIG. 16A). The simple equipment statistical model 550 is a statistical model in which the results of reinforcement learning (constants of the plant model 520, control constants) are applied to an arbitrary simple equipment model 350 (see FIG. 11). For example, the operation parameter determination unit 140 determines a model input that produces the highest output (highest efficiency) as indicated by reference numeral 551 in the relationship between the model input and the model output as indicated by the simplified equipment statistical model 550. In addition, in FIG. 16B, the "model" in the model input and model output refers to the simplified equipment model 350.

[0124] Then, the operation parameter determination unit 140 refers to an operation control map 590 that indicates the operation control of the equipment for the model input. Then, the operation parameter determination unit 140 calculates an operation control value corresponding to a model input value 591 that is most efficient in the simple equipment statistical model 550. Note that in FIG. 16B, the horizontal axes of the simple equipment statistical model 550 and the operation control map 590 are not aligned.

[0125] FIG. 17 is a diagram showing an example of the control parameters. In the example of the control parameters shown in Figure 17, No., model input, model output, simple equipment model output in the simple equipment statistical model 550, control constants as a result of the reinforcement learning process (see Figure 2), and operating parameters as a result of the operating parameter determination process (see Figure 2) are stored. No. is a number that is uniquely assigned to the simplified device model 350 in the plant model 520 . The model input is a numerical value input to the simplified equipment model 350, and the model output is a numerical value output from the simplified equipment model 350 (see FIG. 11). The model input may match the condition-specific model input data 521 in FIG. 16A, and the model output may match the condition-specific model output data 522 in FIG. 16A. In the column of the simple equipment model output for the simple equipment statistical model 550, the output when the simple equipment model 350 is applied to the simple equipment statistical model 550 is stored.

[0126] In the column of the control constant in the reinforcement learning process, the optimal control constant estimated by reinforcement learning is stored for each simplified device model 350. The driving parameters in the driving parameter determination process are stored as driving parameters output in the driving parameter determination process (S401 in FIG. 16A).

[0127] [Simulation process (Step S5 in Figure 2)] (Driving simulator system 2) Fig. 18 is a diagram showing an example of a driving simulator system 2 using a plant model 520 generated as a result of this embodiment. The driving simulator system 2 shown in Fig. 18 is used in step S5 (simulation process) in Fig. 2. As shown in FIG. 18, the driving simulator system 2 has a training operation terminal 21 , a training unit computer 22 , a training controller 23 , a maintenance tool 24 , and a model computer 25 . The training operation terminal 21 is a terminal for operating the plant model 520 when an operation simulation is performed by a user. The training unit computer 22 reflects the operation of the plant model 520 by the training operation terminal 21 in the plant model 520 . The training controller 23 executes the control logic in the driving simulation. The maintenance tool 24 controls the maintenance simulation for the plant model 520 . The model calculator 25 executes the operation (driving simulation) of the plant model 520. Specifically, the driving simulator system 2 (digital twin) is constructed by the instructor P setting control constants, operating parameters, operating sequences, etc. as initial values ​​in the plant model 520 stored in the model calculator 25. Furthermore, the instructor P sets the driving data 301 (see FIG. 2) and the simulated driving data 302 as initial values ​​in the driving simulator system 2, thereby starting the driving simulation.

[0128] In this manner, in the driving simulator system 2, a simulation of the plant is performed using the plant model 520 in which the estimated control constants are set.

[0129] If the results of the operation simulation by the operation simulator system 2 are satisfactory, the user reflects the control constants, operation parameters, and operation sequence in the actual plant. If the results of the driving simulation by the driving simulator system 2 are not good, the process returns to step S1 in Fig. 2. This process corresponds to step S6 in Fig. 2.

[0130] (Comparative Example) FIG. 19 is a diagram showing an example of a driving simulator system 2A according to a comparative example. As shown in FIG. 19, the configuration of a driving simulator system 2A is similar to that of the driving simulator system 2 shown in FIG. In the conventional operation simulator system 2A, operators are trained by simulating the state at the time of an accident that is set in advance as a malfunction (operation to respond to an accident). In the conventional operation simulator system 2A, the user needs to set the equipment model of each system in the plant (main steam system, main steam turbine, condensate system, etc.) in detail / simply one by one using a program. In contrast, in the operation simulator system 2 of this embodiment, as described above, a simplified equipment model 350 (see FIG. 11) is generated based on a performance calculation formula. As a result, the plant simulator system Z of this embodiment can construct an operation simulator system 2 that simulates a plant with simple operations.

[0131] Hereinafter, the driving simulator system 2 of this embodiment will be compared with a conventional driving simulator system 2A with reference to FIGS. The driving simulator system 2 uses the operation data 301 obtained in FIG. 16A and each constant including the control constant as the initial value, and performs training for performing highly efficient operation. As described above, the main purpose of the conventional driving simulator system 2A is to perform an operation to respond to an accident (malfunction). In contrast, the driving simulator system 2 using the plant model 520 generated in this embodiment uses the simplified equipment model 350 based on the performance calculation formula, and further sets the control constant for achieving highly efficient operation by reinforcement learning. By using the driving simulator system 2 with such settings, it is possible to know what settings should be made to the equipment constituting the plant in order to perform highly efficient operation. That is, the driving simulator system 2 enables training for performing highly efficient operation. Of course, as with the conventional driving simulator system 2A, it is possible to train in a state where the plant is not operating. That is, the plant model 520 generated in this embodiment allows parameters that are difficult to set in the actual plant to be tried.

[0132] The plant simulator system Z in this embodiment generates a digital twin of a plant such as a thermal power plant or a hydroelectric power plant based on a performance calculation formula for calculating the performance of equipment (used in plant performance diagnosis).

[0133] Furthermore, the plant simulator system Z in this embodiment is capable of searching and displaying the hierarchical structure of the calculation items in the performance calculation formula, and analyzing factors regarding the operating state of the plant according to the operating level. The plant simulator system Z in this embodiment can test the performance of the plant by testing the operation of the equipment of the plant under predetermined conditions.

[0134] Also, from an educational point of view, it is possible to train the load adjustment capability in the future energy mix and the operation ability required for highly efficient operation in the situation where supply and demand balance is required. As described above, similarly to the operation simulator system 2A, it is also possible to train how to respond to possible abnormal signs and prediction of progression, etc., using setting conditions that are not executed in the actual plant. In this way, the operation simulator system 2 of this embodiment makes it possible to simulate conditions that are difficult to implement in the actual plant.

[0135] As described above, in order to simulate a state that is difficult to implement in an actual plant, the conventional operation simulator system 2A has responded by generating an equipment model for each piece of equipment. However, this equipment model requires programming each of the various conditions, such as the main engine, auxiliary engine, plant configuration, and machine characteristics. In the plant model 520 generated in this embodiment, the various conditions, such as the main engine, auxiliary engine, plant configuration, and machine characteristics, are obtained by reinforcement learning based on a performance calculation formula. Therefore, it is not necessary to program each of the various conditions, such as the main engine, auxiliary engine, plant configuration, and machine characteristics.

[0136] Specifically, as described above, the plant simulator system Z according to this embodiment has the following features: (E1) Pseudo driving data 302 is generated from driving data 301 by a statistical model 311, and data interpolation is performed using this pseudo driving data 302. (E2) A simplified equipment model 350, which is a model of the equipment, is generated based on the performance calculation formula. (E3) Simulation data is generated by a plant model 520 that is combined with the simplified device model 350. By using this method, it is not necessary to program a wide variety of conditions such as the main engine, auxiliary machinery, plant configuration, machine characteristics, etc. one by one. In other words, it is possible to generate the plant model 520 from the operation data 301 without programming a wide variety of conditions such as the main engine, auxiliary machinery, plant configuration, machine characteristics, etc. one by one.

[0137] In these methods, the generation process of the plant model 520 does not become a black box by displaying the visualization display area 402 shown in Fig. 9. That is, the visualization display area 402 shows a hierarchical structure of the calculation symbols used in the performance calculation formula. Therefore, the input / output relationship of the simplified device model 350 generated from the performance calculation formula can also be visually confirmed in the visualization display area 402. Therefore, by displaying the visualization display area 402, the generation process of the plant model 520 does not become a black box. Also, by displaying the operation levels in the operation level determination area 401, the operating state of the actual plant can be easily confirmed.

[0138] Furthermore, the performance calculation formula and the generated plant model 520 may be visualized (displayed on the display device 14). This makes it possible to easily change the simplified device model 350 even when changing the configuration of the actual plant.

[0139] In short, the driving simulator system 2 according to this embodiment has an accident response operation (malfunction) function, and enables driving training using control constants that result in highly efficient driving, as well as the state at the time of an accident. Furthermore, the operation simulator system 2 according to this embodiment allows training in operating conditions that have not been performed in an actual plant until now. This makes it easy to train in countermeasure support for abnormality signs and progression prediction.

[0140] For example, a user can try reducing fuel consumption and CO2 emissions by exploring various setting values ​​in a driving simulation using the driving simulator system 2. In addition, it is also possible to obtain highly efficient control constants while reducing CO2 emissions through reinforcement learning.

[0141] In this way, in this embodiment, a digital twin cycle is run in which the plant (thermal power plant) is reproduced and verified in a virtual space by the operation simulator system 2. This makes it possible to analyze the causes of a decrease in power generation efficiency due to measures to phase out inefficient coal in 2030, verification of combustion methods such as biomass co-firing, by-product co-firing, ammonia co-firing, and hydrogen co-firing, and adjustment power operation.

[0142] In other words, by utilizing the plant simulator system Z in this embodiment, business operators can not only simulate plant operation in biomass co-firing, by-product co-firing, ammonia co-firing, and hydrogen co-firing, but also analyze the factors at the time of power generation efficiency reduction due to adjustment power operation. In addition, business operators can set a target value for power generation efficiency (41.0% or more for coal-fired power generation) that is consistent with the energy mix for fiscal year 2030, and continue business activities while maximizing profits and complying with environmental regulations.

[0143] Furthermore, for the reasons described above, the plant simulator system Z of this embodiment is expected to provide thermal, hydroelectric, and renewable energy power generation companies with not only increased electricity sales revenue, but also reduced fuel costs and improved equipment operating rates through highly efficient plant operation, as well as reduced costs for education and training in power plant operation. It is also possible to grasp the plant status and improve the accuracy of the plant model 520, realize highly reliable monitoring, operation, control, and recovery of the power plant, and predict the progress of unexpected projects and consider countermeasures. This makes it possible to transfer the know-how of skilled workers from on-site training through OJT to a digital power plant, enabling the construction, evaluation, and confirmation of flexible operation methods.

[0144] It is expected that it will become difficult to grasp the state of the grid due to the separation of power generation and transmission, the increase in renewable energy sources, the activation of consumer-side equipment, and the like. As a result, there is a risk that it will become difficult to collect real-time data on the amount of power demand. In order to realize economical and reliable power generation operation, the importance of simulation technology and performance evaluation that can accurately predict and evaluate the behavior of power demand will increase, and the accuracy of the model constants will become more important. In this embodiment, the hierarchical structure of the performance calculation formula and the operation level are derived and displayed while utilizing the statistical model 311 from the actual data of the thermal power plant. Furthermore, in this embodiment, the control constants are estimated based on the performance calculation formula. Then, the operation sequence that realizes the control constants, etc. are output. By adding elements that take into account the progress prediction and countermeasure support for abnormal signs and unexpected events to the training scenario based on the operation method sequence, a plant performance diagnosis that includes digital twin technology that realizes a highly efficient plant that has not been available in the operation simulator system 2A until now is realized.

[0145] Furthermore, the plant simulator system Z of this embodiment has the following advantages. (F1) Understanding the plant status and achieving highly efficient control constants in the plant. (F2) Achieving highly reliable power plant monitoring, operation, control, and recovery. (F3) Predicting progress and considering countermeasures for unexpected business situations. (F4) Construction, evaluation, and confirmation of flexible operation methods according to needs that are not limited by conventional power supply methods. (F5) Evaluate and verify new power plant operation methods on a computer. (F6) Education and training for power plant operations.

[0146] Furthermore, the plant simulator system Z of this embodiment provides the following economic advantages. (F7) The usefulness of plant equipment information and the necessity of stabilization control, etc. can be confirmed qualitatively and quantitatively. (F8) Maximize the use of plant facilities (improve utilization rate). (F9) Maximum minimization of generation constraints in expanding electricity trading.

[0147] The present invention is not limited to the above-described embodiment, and includes various modified examples. For example, the above-described embodiment has been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to all of the configurations described. In addition, it is possible to add, delete, or replace part of the configuration of the described embodiment with other configurations.

[0148] In addition, the above-mentioned configurations, functions, units 110, 120, 121, 122, 130, 131, 132, 140, 150, storage device 12, etc. may be realized in hardware by designing some or all of them as integrated circuits, for example. In addition, as shown in Fig. 1, the above-mentioned configurations, functions, etc. may be realized in software by a processor such as CPU 11 interpreting and executing a program that realizes each function. Information such as a program, table, file, etc. that realizes each function can be stored in a recording device such as memory 100 or SSD (Solid State Drive), or a recording medium such as an IC (Integrated Circuit) card, SD (Secure Digital) card, or DVD (Digital Versatile Disc). In addition, in each embodiment, the control lines and information lines are those that are considered necessary for the explanation, and not all control lines and information lines in the product are necessarily shown. In reality, it can be considered that almost all components are connected to each other. [Explanation of symbols]

[0149] 1. Simulator generator 2 Driving simulator system (simulator section) 2A Driving simulator system 14 Display device (display section) 110 Statistical Processing Unit 120 Analysis Support Processing Unit 121 Hierarchical structure search unit (hierarchical structure processing unit) 122 Analysis support display processing unit (hierarchical structure processing unit, action level processing unit) 130 Control constant estimation unit 131 Plant model processing unit (plant model generation unit) 132 Reinforcement learning unit (control constant estimation unit) 140 Operation parameter determination unit 150 Operation sequence determination unit 301 Operational Data 302 Simulated driving data 304 Control constant information (control constant) 305 Operation parameter information 306 Operation sequence information 311 Statistical Models 320 Performance calculation formula information (performance calculation formula) 350 Simple Equipment Model (Equipment Model) 400 Formula analysis screen 401 Motion level judgment area 402 Visualization display area 520 Plant Model Z Plant Simulator System (Simulator System) S5 Simulation process (simulation step) S321 Reinforcement learning process (control constant estimation step) S311 Generation of simple equipment model (Plant model generation step) S312 Generate input / output list (Plant model generation step)

Claims

1. a plant model generation unit that generates an equipment model that is a model of an equipment constituting a plant based on a performance calculation formula for calculating performance of the equipment, and generates a plant model that is a model of the plant by connecting the equipment models; a control constant estimation unit that estimates a control constant, which is a constant used in the performance calculation formula, by reinforcement learning based on the plant model; a simulator unit that performs a simulation of the plant using the plant model in which the estimated control constants are set; a hierarchical structure processing unit that extracts a hierarchical structure of calculation symbols used in the performance calculation formula based on the performance calculation formula and displays the hierarchical structure on a display unit; A simulator system comprising:

2. an action level processing unit that calculates an action level that is an index for determining an operational state of the plant, and displays the calculated action level together with the hierarchical structure on the display unit; 2. The simulator system according to claim 1, further comprising:

3. The action level processing unit includes: Calculating the operation level based on a best value of the device, which is calculated based on the performance calculation formula and is the output when the operation efficiency of the device is the best, and an operating value obtained from the device; The operating value is The current value of the operating data acquired from the device, or a comparison value including a difference between the operating data at a predetermined time and the current value.

3. The simulator system according to claim 2.

4. a hierarchical structure processing step of extracting a hierarchical structure of calculation symbols used in a performance calculation formula for calculating the performance of equipment constituting a plant, based on the performance calculation formula, and displaying the hierarchical structure on a display unit; a plant model generating step of generating an equipment model that is a model of the equipment based on the performance calculation formula, and generating a plant model that is a model of the plant by connecting the equipment models; a control constant estimation step of estimating a control constant, which is a constant used in the performance calculation formula, by reinforcement learning based on the plant model; a simulation step of simulating the plant using the plant model in which the estimated control constants are set; A simulator method comprising:

Images