Control device, evaluation method, and evaluation system

Abstract

[Problem] To provide a display device, an evaluation method, and an evaluation system which make it possible to easily obtain an appropriate evaluation model. [Solution] Provided is a display device 60 comprising a display screen D1. In a first display area AR1 of the display screen D1, a plurality of evaluation models are configured in a selectable and displayable manner. In a second display area AR2, the results of an evaluation of evaluation data by a selected first evaluation model are configured in a displayable manner. In a third display area AR3, the results of an evaluation of evaluation data by a selected second evaluation model are configured in a displayable manner.

Description

[Technical Field]

[0001] The present invention relates to a display device, an evaluation method, and an evaluation system. [Background technology]

[0002] Conventionally, process data has been acquired from a plurality of sensors installed in a controlled object such as a plant, and the operating status of the controlled object has been monitored and controlled based on the acquired process data. An evaluation model may be used to monitor the operating status of the controlled object based on the process data. The process data acquired from the controlled object is evaluated according to a predetermined evaluation model, and based on the evaluation result, it becomes possible to determine whether the controlled object is operating normally.

[0003] Accurate monitoring of the operating status of controlled objects requires a high-performance evaluation model. Obtaining a high-performance evaluation model requires trial and error, as it involves adjusting the parameters that make up the evaluation model and comparing and selecting multiple evaluation models.

[0004] Patent Document 1 discloses a model information display method for efficiently comparing multiple evaluation models. According to this model information display method, when multiple evaluation models are selected, the multiple evaluation models are simultaneously displayed so that the amount of information displayed varies depending on the number selected. This makes it possible to efficiently compare multiple evaluation models. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Publication No. 2018-206283 Summary of the Invention [Problem to be solved by the invention]

[0006] However, the model information display method described in Patent Document 1 merely provides a display method that enables comparison of multiple evaluation models, and therefore does not sufficiently streamline the work required to obtain an appropriate evaluation model.

[0007] SUMMARY OF THE INVENTION It is therefore an object of the present invention to provide a display device, an evaluation method, and an evaluation system that are capable of efficiently acquiring an appropriate evaluation model. [Means for solving the problem]

[0008] The present disclosure provides a display device having a display screen configured to allow a plurality of evaluation models to be selectably displayed in a first display area, to allow an evaluation result of evaluation data based on the selected first evaluation model to be displayed in a second display area, and to allow an evaluation result of evaluation data based on the selected second evaluation model to be displayed in a third display area.

[0009] The evaluation data is data evaluated by the evaluation model. For example, process data (including raw data that has not been subjected to computational processing) acquired from a plurality of sensors installed in a controlled object such as a plant can be evaluated as the evaluation data. The process data may vary depending on the season and other environmental factors even when controlled based on the same control data. Furthermore, the evaluation data may vary due to changes over time in the components of the controlled object even when controlled based on the same control data.

[0010] Furthermore, the fourth display area of ​​the display screen may be configured to be able to set and display a first evaluation criterion for evaluation using a first evaluation model, and the fifth display area of ​​the display screen may be configured to be able to set and display a second evaluation criterion for evaluation using a second evaluation model, the second display area may be configured to be able to display the evaluation results of the evaluation data using the selected first evaluation model based on the set first evaluation criterion, and the third display area may be configured to be able to display the evaluation results of the evaluation data using the selected second evaluation model based on the set second evaluation criterion.

[0011] Furthermore, the sixth display area of ​​the display screen may be configured to be able to set and display the period of the evaluation data, the second display area may be configured to be able to display the evaluation results of the evaluation data for the set period using the first evaluation model, and the third display area may be configured to be able to display the evaluation results of the evaluation data for the set period using the second evaluation model. The period of evaluation data refers to the period during which the evaluation data was acquired.

[0012] The first evaluation model is generated based on a training data range (a plurality of training data belonging to the predetermined range) by a machine learning model having predetermined parameters. The display device may further be configured to configurably display the parameter values ​​and the training data range in a seventh display area of ​​the display screen, and to display an evaluation result of the evaluation data by the evaluation model generated by the machine learning model having the set parameter values ​​in the second display area. Note that at least a portion of the parameter values ​​or the training data range may be configurably displayed in the seventh display area within the same display screen that displays the first to sixth display areas, or a transition may be made to a different display screen via the seventh display area, where the parameter values ​​or the training data range may be configurably displayed. In the former case, it is possible to change the parameter values, etc. in the seventh display area and then display the display result in the second display area, thereby making it possible to efficiently obtain an appropriate evaluation model.

[0013] Here, the machine learning model includes those that use neural networks such as convolutional neural networks (CNNs), those that use regression models such as Gaussian process regression, and those that use tree algorithms such as decision trees.

[0014] The learning data is also called teacher data. The learning data includes, for example, process data when a controlled object such as a plant is operating smoothly. The learning data also includes, for example, process data when a controlled object such as a plant is operating smoothly at high power, process data when the controlled object is operating smoothly at medium power, and process data when the controlled object is operating smoothly at low power.

[0015] Machine learning parameters include configurable variables. When the parameters are different, different evaluation models are generated based on the same training data. Parameters include hyperparameters, such as the number of layers in a decision tree, the number of units in the hidden layer in a neural network, and a parameter indicating weight decay.

[0016] The first display area may be configured to allow machine learning models having set parameter values ​​to be selectably displayed in the first display area together with a plurality of evaluation models. The first display area may be configured to display at least one of information indicating that the evaluation model is an evaluation model that has been used in the past to evaluate process data acquired from a controlled object, and information indicating that the evaluation model is an evaluation model that is currently being used to evaluate process data acquired from a controlled object, among the multiple evaluation models. Furthermore, the first display area may be configured to display an evaluation model selected from the plurality of evaluation models in a manner that allows copying or deletion.

[0017] The present disclosure provides an evaluation method, which includes the steps of: displaying a plurality of selectable evaluation models in a first display area of ​​a display screen; displaying an evaluation result of the evaluation data using the selected first evaluation model in a second display area of ​​the display screen; and displaying an evaluation result of the evaluation data using the selected second evaluation model in a third display area of ​​the display screen.

[0018] The present disclosure provides a rating system, comprising: means for selecting a plurality of rating models; means for obtaining a rating result of rating data according to a selected first rating model; and means for obtaining a rating result of rating data according to a selected second rating model.

[0019] The present disclosure provides an evaluation system for evaluating an operational status of a plant. The evaluation system includes means for selecting a first evaluation model generated using process data acquired from a sensor installed in the plant at a first time period as learning data and a second evaluation model generated using process data acquired from a sensor installed in the plant at a second time period as learning data. The selected first evaluation model evaluates the evaluation data acquired from the sensor installed in the plant. The evaluation results are configured to be displayed. The selected second evaluation model evaluates the same evaluation data acquired from the sensor installed in the plant. The evaluation results are configured to be displayed. The evaluation data may include process data when the operational status of the plant becomes abnormal. The first evaluation model may be generated based on the learning data by a machine learning model having one or more parameters set to predetermined values. The one or more parameters may be configured to be configurable. When a parameter is newly set, a new evaluation model may be generated by the machine learning model having the set parameter.

[0020] The present disclosure provides an evaluation method for evaluating the operating status of a plant. The evaluation method includes a step of selectably displaying, in a first display area of ​​a display screen, a first evaluation model generated using process data acquired from a sensor installed in the plant at a first time period as learning data, and a second evaluation model generated using process data acquired from a sensor installed in the plant at a second time period as learning data. The method further includes a step of displaying, in a second display area of ​​the display screen, an evaluation result of the evaluation data using the selected first evaluation model, and a step of displaying, in a third display area of ​​the display screen, an evaluation result of the evaluation data using the selected second evaluation model. The method further includes a step of evaluating the second evaluation data based on either the selected first evaluation model or the selected second evaluation model. [Brief explanation of the drawings]

[0021] [Figure 1] Schematic diagram of Plant 1 [Figure 2] Functional block diagram of evaluation system 30 [Figure 3] Display screen of the display device 60 [Figure 4] 1 is a block diagram showing the physical configuration of the evaluation system 30. [Figure 5] Flowchart of the evaluation method according to this embodiment DETAILED DESCRIPTION OF THE INVENTION

[0022] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The following embodiments of the present invention will be described with reference to the accompanying drawings. The following embodiments are merely examples for explaining the present invention, and are not intended to limit the present invention to these embodiments. The controlled objects to which the present invention is applicable include plants. Examples of plants include power plants including boilers, incineration plants, chemical plants, wastewater treatment plants, and other plants from which process data can be acquired. Note that the process data includes unprocessed data acquired from sensors and the like.

[0023] FIG. 1 is a schematic diagram showing the overall configuration of a plant 1 according to this embodiment. The plant 1 according to this embodiment is a power plant equipped with a circulating fluidized bed boiler (Circulating Fluidized Bed type) that generates steam by burning fuel while circulating a circulating material such as silica sand that flows at high temperature. As fuel for the plant 1, in addition to fossil fuels such as coal, non-fossil fuels (woody biomass, waste tires, waste plastic, sludge, etc.) can be used. The steam generated in the plant 1 is used to drive a turbine 100.

[0024] The plant 1 is configured to combust fuel in a furnace 2, separate circulating material from exhaust gas using a cyclone 3 that functions as a solid-gas separator, and return the separated circulating material to the furnace 2 for circulation. The separated circulating material is returned to the bottom of the furnace 2 via a circulating material recovery pipe 4 connected below the cyclone 3. The bottom of the circulating material recovery pipe 4 is connected to the bottom of the furnace 2 via a loop seal 4a with a narrowed flow path. This leaves a predetermined amount of circulating material stored in the bottom of the circulating material recovery pipe 4. The exhaust gas from which the circulating material has been removed by the cyclone 3 is supplied to the rear flue 5 via the exhaust gas flow path 3a.

[0025] The boiler includes a furnace 2 for burning fuel and a heat exchanger for generating steam and the like using heat obtained by combustion. A fuel supply port 2a for supplying fuel is provided in the middle of the furnace 2, and a gas outlet 2b for discharging combustion gas is provided in the upper part of the furnace 2. Fuel supplied to the furnace 2 from a fuel supply device (not shown) is supplied into the furnace 2 through the fuel supply port 2a. In addition, a furnace wall tube 6 for heating boiler feedwater is provided on the furnace wall of the furnace 2. The boiler feedwater flowing through the furnace wall tube 6 is heated by combustion in the furnace 2.

[0026] Within the furnace 2, solids containing fuel supplied from the fuel supply port 2a are fluidized by combustion and fluidization air introduced from the lower air supply line 2c, and the fuel burns at, for example, approximately 800 to 900°C while flowing. Combustion gas generated in the furnace 2 is introduced into the cyclone 3, accompanied by circulating material. The cyclone 3 separates the circulating material from the gas by centrifugal separation, and returns the separated circulating material to the furnace 2 via the circulating material recovery pipe 4, while sending the combustion gas from which the circulating material has been removed through the exhaust gas flow path 3a to the rear flue 5.

[0027] In the furnace 2, a portion of the circulating material, called the in-furnace bed material, accumulates at the bottom. This bed material may contain coarse-grained bed material or exhaust combustion impurities that are unsuitable for circulating flow. These unsuitable bed materials can cause poor flow. To prevent this, the in-furnace bed material is continuously or intermittently discharged to the outside from a discharge port 2d at the bottom of the furnace 2. After removing impurities such as metals and coarse grains from the discharged bed material via a circulation line (not shown), the discharged bed material is either fed back to the furnace 2 or discarded as is. The circulating material in the furnace 2 circulates within a circulation system consisting of the furnace 2, cyclone 3, and circulating material recovery pipe 4.

[0028] The rear flue 5 has a flow path for flowing the gas discharged from the cyclone 3 to a subsequent stage. The rear flue 5 has a superheater 10 that generates superheated steam and an economizer 12 that preheats boiler feedwater as an exhaust heat recovery section that recovers heat from the exhaust gas. The exhaust gas flowing through the rear flue 5 is cooled by heat exchange with the steam and boiler feedwater flowing through the superheater 10 and the economizer 12. The rear flue 5 also has a steam drum 8 that stores the boiler feedwater that has passed through the economizer 12, and the steam drum 8 is also connected to the furnace wall tubes 6.

[0029] The economizer 12 transfers heat from the exhaust gas to the boiler feedwater to preheat the boiler feedwater. The economizer 12 is connected to the pump 7 by a pipe 21, and is also connected to the steam drum 8 by a pipe 22. The boiler feedwater is supplied from the pump 7 via the pipe 21 to the economizer 12 and preheated by the economizer 12, and is then supplied to the steam drum 8 via the pipe 22.

[0030] A downcomer pipe 8a and a furnace wall pipe 6 are connected to the steam drum 8. The boiler feedwater in the steam drum 8 flows down the downcomer pipe 8a, is introduced into the furnace wall pipe 6 at the bottom of the furnace 2, and flows toward the steam drum 8. The boiler feedwater in the furnace wall pipe 6 is heated by the combustion heat generated in the furnace 2, and evaporates into steam in the steam drum 8.

[0031] A saturated steam pipe 8b that discharges the steam inside is connected to the steam drum 8. The saturated steam pipe 8b connects the steam drum 8 and a superheater 10. The steam inside the steam drum 8 is supplied to the superheater 10 via the saturated steam pipe 8b. The superheater 10 superheats the steam using the heat of the exhaust gas to generate superheated steam. The superheated steam passes through a pipe 10a and is supplied to a turbine 100 outside the plant 1 and is used for power generation.

[0032] The pressure and temperature of the steam discharged from the turbine 100 are lower than the pressure and temperature of the steam discharged from the superheater 10. Although not particularly limited, the pressure of the steam supplied to the turbine 100 is approximately 10 to 17 MPa, and the temperature is approximately 530 to 570°C. The pressure of the steam discharged from the turbine 100 is approximately 3 to 5 MPa, and the temperature is approximately 350 to 400°C.

[0033] A condenser 102 is provided downstream of the turbine 100. The steam discharged from the turbine 100 is supplied to the condenser 102, where it is condensed and returned to saturated water, and then supplied to the pump 7. A generator is connected to the turbine 100, which converts the kinetic energy obtained by the rotation of the turbine 100 into electrical energy.

[0034] The pump 7a supplies makeup water so as to maintain a constant water level in the condenser 102. Fig. 1 shows a flow rate u1 of makeup water supplied by the pump 7a (an example of "process data").

[0035] The process data handled in this embodiment may be any data related to the plant 1. For example, it may be data (an example of "process data") measuring the state of the plant 1 with a sensor. More specifically, it may include measured values ​​of the temperature, pressure, flow rate, and the like of the plant 1. FIG. 1 shows a boiler feedwater flow rate u2 (an example of "process data") supplied from the pump 7 to the economizer 12. FIG. 1 also shows a boiler outlet steam flow rate u3 (an example of "process data") supplied from the superheater 10 to the turbine 100, and a saturated steam flow rate u4 (an example of "process data") supplied from the steam drum 8 to the superheater 10. The make-up water flow rate u1 may be controlled to follow the saturated steam flow rate u4. The boiler feedwater flow rate u2 may be controlled to be adjusted while monitoring both the boiler outlet steam flow rate u3 (or the superheated steam flow rate) and the liquid level in the steam drum 8.

[0036] If a hole occurs in the pipe system that makes up plant 1, the makeup water flow rate u1 will increase, and the flow rate difference between the boiler feedwater flow rate u2 and the boiler outlet steam flow rate u3 will increase. A DCS (Distributed Control System, Figure 2) 20 receives process data for plant 1, such as the makeup water flow rate u1, boiler feedwater flow rate u2, boiler outlet steam flow rate u3, and saturated steam flow rate u4, from plant 1, and monitors the operating status of plant 1 to check for any abnormalities in plant 1.

[0037] Although the makeup water flow rate u1, the boiler feedwater flow rate u2, the boiler outlet steam flow rate u3, and the saturated steam flow rate u4 are exemplified as process data, other data may be used as the process data for the plant 1. The process data for the plant 1 may be other data such as temperature and pressure, or data calculated based on multiple process data.

[0038] FIG. 2 is a diagram showing functional blocks of the evaluation system 30 according to this embodiment. The DCS 20 is a distributed control system for controlling the plant 1. The DCS 20 acquires process data from sensors and the like installed in the plant 1, and supplies control signals to the plant 1 for controlling the plant 1 based on the process data.

[0039] The evaluation system 30 includes an evaluation model management device 40 for managing and generating an evaluation model for evaluating the operating status of the plant 1, a plant 1 monitoring device 50 for acquiring process data from the DCS 20 and monitoring the operating status of the plant 1 based on a selected predetermined evaluation model, and a display device 60 for displaying images for managing and generating the evaluation model by the evaluation model management device 40 and the operating status of the plant 1 monitored by the plant 1 monitoring device 50. The components constituting the evaluation system 30 do not need to be configured integrally. For example, some or all of the evaluation model management device 40 may be provided in a remote location and communicatively connected to the other components via a communication network such as the Internet.

[0040] The evaluation model management device 40 includes a storage unit 42 and a model selection unit 44 . The storage unit 42 includes a process data storage unit 42A, a cleansing rule storage unit 42B, and a trained model storage unit 42C. The process data storage unit 42A stores process data acquired from the DCS 20. The process data may include, for example, process data of the plant 1 or other plants over a long period of time. The process data includes learning data (teaching data) for generating an evaluation model. The learning data may include, for example, process data when a controlled object, such as the plant 1, is operating at high power, process data when it is operating well at medium power, and process data when it is operating well at low power. The learning data may include process data when the controlled object is operating well, i.e., to achieve a predetermined power, and process data when the controlled object has generated an abnormality. The process data may be stored in association with time information when the process data was acquired. Furthermore, the process data may be stored in association with the power output and the state of good operation when the process data was acquired. Note that the process data does not necessarily have to be acquired via the DCS 20 and may be acquired directly from the plant 1, for example.

[0041] The cleansing rule storage unit 42B stores rules for performing data cleansing to extract learning data from process data. For example, when it is desired to extract learning data when the operating status of the plant 1 is good, the cleansing rule storage unit 42B stores rules for performing cleansing to remove data that may lead to performance degradation.

[0042] The trained model storage unit 42C stores multiple trained models generated based on training data using a machine learning model. The machine learning model is, for example, a Gaussian process regression model. The training data is, for example, three types of process data from the same period in January 2020: process data when plant 1 or a similar plant was operating well at high power, process data when plant 1 was operating well at medium power, and process data when plant 1 was operating well at low power. The trained model storage unit 42C uses this process data as training data and stores a trained model generated based on a predetermined machine learning model as one of the evaluation models. Similarly, the trained model storage unit 42C stores multiple trained models as evaluation models. The physical configuration of the storage unit 42 will be described later.

[0043] The model selection unit 44 selects an evaluation model to be used for evaluating the process data acquired from the DCS 20 . The learning period registration unit 44A of the model selection unit 44 registers and stores the period of process data extracted as learning data. An operator can register the learning period from the display screen of the display device 60. For example, if the periods during which the plant operated well at high output, medium output, and low output are known in advance, the operator can register these periods as learning periods. The model learning unit 44C can generate a trained model using the process data acquired during the period registered in the learning period registration unit 44A as learning data.

[0044] The cleansing unit 44B executes a cleansing process on the process data. Specifically, the cleansing process is executed on the process data for a predetermined period extracted by the learning period registration unit 44A based on the cleansing rules stored in the cleansing rule storage unit 42B, and the process data that is not suitable for generating an evaluation model by performing predetermined machine learning is removed.

[0045] The model learning unit 44C performs machine learning according to a predetermined machine learning model based on the process data that is learning data, generates an evaluation model, and stores it in the learned model storage unit 42C. Specifically, the model learning unit 44C acquires the process data acquired during the period registered in the learning period registration unit 44A from the process data storage unit 42A, and performs machine learning on the process data that has been cleansed by the cleansing unit 44B as learning data to generate an evaluation model.

[0046] The model selection unit 44D accepts the selection of an evaluation model by the operator and selects the evaluation model to be used for evaluation by the model evaluation unit 44G. The model selection unit 44D is capable of simultaneously selecting, for example, two or more evaluation models. The model selection unit 44D displays the multiple evaluation models stored in the trained model storage unit 42C on the display screen of the display device 60 so that they can be selected. The operator can select multiple evaluation models to be used for evaluation from the evaluation models displayed on the display screen.

[0047] The evaluation period registration unit 44E registers and stores the period of process data extracted as evaluation data. The evaluation period registration unit 44E displays the evaluation period on the display screen of the display device 60 so that it can be set. The operator can register the evaluation period from the display screen of the display device 60. The model evaluation unit 44G evaluates the process data acquired during the period registered in the evaluation period registration unit 44E as evaluation data.

[0048] The threshold registration unit 44F accepts a threshold (an example of an "evaluation criterion") set by the operator, and registers and stores the threshold as an evaluation criterion for evaluation by the model evaluation unit 44G. The threshold registration unit 44F can register and store a threshold set for each evaluation model by the model evaluation unit 44G. The threshold registration unit 44F displays a settable threshold on the display screen of the display device 60 for each evaluation model selected by the model selection unit 44D. The operator can register a threshold for each evaluation model from the display screen of the display device 60.

[0049] The model evaluation unit 44G evaluates the evaluation data according to the evaluation model. Specifically, the process data acquired during the evaluation period registered in the evaluation period registration unit 44E is input as evaluation data to multiple evaluation models selected by the model selection unit 44D, and an evaluation result is obtained. If a threshold is registered in the threshold registration unit 44F, the model evaluation unit 44G uses it as an evaluation criterion for obtaining the evaluation result.

[0050] The evaluation result display unit 44H displays the evaluation results of the evaluation data according to the selected evaluation model on the display screen of the display device 60 for each evaluation model.

[0051] The monitoring device 50 evaluates the process data acquired from the DCS 20 in accordance with the finally selected evaluation model, thereby monitoring the plant 1. For example, the monitoring device 50 can evaluate the process data currently acquired from the DCS 20 in accordance with the finally selected evaluation model and a set threshold, and display an alarm or the like on the display screen of the display device 60 when process data exceeding the threshold is acquired.

[0052] The display device 60 receives control signals (including data) for displaying the above-mentioned various information from the evaluation model management device 40 and the monitoring device 50, and displays them on the display screen. Figure 3 shows an example of a display screen D1 of the display device 60.

[0053] As shown in the figure, a plurality of evaluation models are displayed in an area AR1 (an example of a "first display area") of the display device 60. The worker can select a plurality of evaluation models from the displayed evaluation models.

[0054] In the figure, nine selectable evaluation models, Model No. 1 to Model No. 8, are displayed in area AR1. An operator can select an evaluation model by checking a checkbox in the "Evaluation Model Selection" column. The figure shows that evaluation models Model No. 4 and Model No. 7 have been selected. The "Learning Date and Time" column indicates the date and time when the corresponding evaluation model was generated. The "Production Operation" column indicates whether the corresponding evaluation model has been used currently or in the past to monitor the plant 1 by the monitoring device 50. The figure shows that the evaluation model Model No. 1 is currently being used to monitor the plant 1 by the monitoring device 50, and the evaluation model Model No. 3 has been used to monitor the plant 1 by the monitoring device 50 in the past. Therefore, the monitoring device 50 monitors the plant 1 using the evaluation model Model No. 1, using the process data acquired from the DCS 20 as evaluation data. In this way, by indicating which evaluation models have been used in the past and which are currently being used to evaluate process data acquired from a controlled object, operators can use these evaluation models as a reference to compare different evaluation models. The "Above Threshold" column and the "Below Threshold" column display the upper and lower thresholds (an example of an "evaluation criterion"), if any, set as evaluation criteria for the corresponding evaluation model. For example, if the value of the process data exceeds the threshold value relative to the reference value, it may be determined that the operating status of plant 1 is not good.

[0055] At the same time, areas AR81 and AR82 on the display screen D1 display the contents of the learning data used to generate the evaluation models No. 7 and No. 4 selected in area AR1. By viewing the contents of the learning data, the operator can determine whether the evaluation model to be selected was generated based on learning data appropriate for evaluating the evaluation data. That is, in a controlled object such as plant 1, whose state fluctuates from moment to moment due to the season, weather, other environmental factors, and changes over time, selecting an evaluation model generated based on learning data similar to the evaluation data to be evaluated can improve evaluation accuracy. For example, it may be inappropriate to use an evaluation model generated based on learning data acquired in winter to evaluate evaluation data obtained in summer. The display device 60 according to this embodiment displays the contents of the learning model used to generate the evaluation model selected in area AR1, thereby increasing the operator's chances of obtaining an appropriate evaluation model.

[0056] Furthermore, the evaluation system 30 can extract process data obtained over multiple discrete periods as training data for generating a single evaluation model. For example, in the figure, a single evaluation model is generated based on process data obtained over three periods, learning period 1 through learning period 3. In other words, for a controlled object such as a plant 1 that can assume multiple states (low output, medium output, and high output), generating an evaluation model using training data that includes each of these states makes it possible to generate a highly versatile evaluation model. Based on the process data storage unit 42A or other information, the operator can confirm that the training data is configured to include process data of the plant 1 that indicates good operating conditions in each state. The display device 60 may also be configured so that the operator can directly input the training period into an area within the display screen D1, thereby setting the training period for the evaluation model to be generated without leaving the display screen D1.

[0057] At the same time, the area AR6 (an example of a "sixth display area") of the display screen D1 displays the evaluation data period, i.e., the evaluation period of the plant 1. By inputting the period, the operator can set the period during which the process data to be evaluated by the evaluation model was acquired. With this configuration, the display device 60 enables the acquisition of an appropriate evaluation model. For example, the operator can determine whether the selected evaluation model is appropriate by checking the evaluation results of multiple evaluation models while varying the evaluation period. For example, by setting the evaluation data period to include the period during which an abnormality occurred in the plant 1, the operator can check the evaluation results of the evaluation model and determine whether the evaluation model can detect the abnormality. Then, for an evaluation model that detected an abnormality, the operator can check whether the evaluation model does not detect the abnormality by setting the evaluation data period to the period during which the plant 1 was operating smoothly and checking the evaluation results of the evaluation model. The display screen D1 is configured to simultaneously display the area AR6 for setting the evaluation data period and the areas AR2 and AR3 for displaying the evaluation results of the evaluation models, allowing the operator to quickly acquire an appropriate evaluation model.

[0058] At the same time, thresholds are displayed in areas AR4 (an example of a "fourth display area") and AR5 (an example of a "fifth display area") on the display screen D1 as examples of evaluation criteria for the evaluation model selected in area AR1. The operator can set thresholds for evaluation using the evaluation model of Model No. 7 by inputting upper and lower threshold limits in area AR4. Similarly, the operator can set thresholds for evaluation using the evaluation model of Model No. 4 by inputting upper and lower threshold limits in area AR5. When the operator inputs a new threshold using the input unit 30 and presses the "Evaluate" button, the evaluation results based on the new, changed thresholds are displayed in area AR2. Therefore, by setting a threshold value that detects an abnormality when the evaluation data period is set to include a time when an abnormality occurred in plant 1, and by setting a threshold value that does not detect an abnormality when the evaluation data period is set to include a time when plant 1 was operating smoothly, it is possible to obtain an appropriate evaluation model and appropriate evaluation criteria. Furthermore, by comparing the evaluation results when the same threshold value is set in areas AR4 and AR5, it is possible to determine the merits or demerits of the selected evaluation model. The threshold value can be configured to be set using an index such as an absolute value, a ratio to a reference value, or a deviation.

[0059] At the same time, an area AR2 (an example of a "second display area") on the display screen D1 displays the evaluation results of the evaluation data using one of the selected evaluation models, and an area AR3 (an example of a "third display area") displays the evaluation results of the evaluation data using the other selected evaluation model. In area AR2, as the evaluation results, an estimated expected value calculated based on the learning data is shown by a solid line, and upper and lower thresholds based on the estimated expected value are shown by dashed lines. The values ​​of multiple evaluation data are also shown discretely. The learning data values ​​may also be shown discretely. In the figure, for example, in a two-dimensional space with the horizontal axis representing the boiler load of the plant 1 (an example of "process data") and the vertical axis representing the temperature difference between the upper and lower furnace temperatures (an example of "process data"), the evaluation results are displayed by showing a region that can be determined as a good operating range, i.e., a region bounded by the upper and lower thresholds including the solid line representing the estimated expected value, and the position of the evaluation data within the same two-dimensional space. However, the manner in which the evaluation results are displayed is not limited to this. For example, the degree of deviation from the optimal operating range may be scored and the evaluation result may be displayed using an index called healthiness. Alternatively, the evaluation result may be displayed using only text, for example. However, as shown in this embodiment, by selecting two process data as learning data and displaying the range of good operating conditions and the position of the evaluation data in a two-dimensional space centered on the two process data as axes so that they can be distinguished, operators can easily understand the evaluation result. In the figure, there is evaluation data that exceeds the good operating range, so according to the evaluation model No. 7, it is determined that an abnormality has occurred in plant 1.

[0060] As shown in the figure, area AR3 shows the evaluation results of evaluation model No. 4 for the same evaluation data. Areas AR2 and AR3 show evaluation results using the same indicators for the same evaluation data, allowing the operator to easily compare the evaluation models. Note that the training data for different evaluation models does not necessarily have to be different; some or all of the data may be the same. For example, by varying some of the parameters, it is possible to generate different evaluation models. Therefore, it is possible to generate different evaluation models based on the same training data.

[0061] Here, when the worker changes the period in area AR6 and presses the "Evaluation" button in area AR4 with an input means such as a mouse, the evaluation results based on the changed evaluation data are displayed in area AR1. Furthermore, when the operator changes the threshold value in area AR4 and presses the "Evaluate" button in area AR4 with an input means such as a mouse, the evaluation result based on the changed threshold value is displayed in area AR1.

[0062] In this way, it is possible to change multiple evaluation models, change evaluation data, change evaluation criteria, and display evaluation results with the changed content within a single display screen D1, allowing the worker to efficiently obtain an appropriate evaluation model.

[0063] Various process data can be selected as the training data and evaluation data. For example, to generate an evaluation model for detecting a specific abnormality, two or more process data correlated with the abnormality can be selected. For example, to detect a plant abnormality known as a "fracture," such as a tube leak, in which the metal material of a boiler tube is damaged and ruptured, causing the steam inside to leak to the outside, the training data and evaluation data can include process data obtained from multiple acoustic emission (AE) sensors installed in the boiler tubes to detect the fracture. Other process data, such as information indicating the temperature of the piping system (including temperature information obtained directly or indirectly from sensors and information indicating the temperature difference at specified locations), information indicating the amount of heat absorption, information indicating the valve opening, and information indicating the temperature of the gas medium, can be acquired. Based on this information, evaluations can be performed simultaneously according to multiple evaluation models. In addition, the present invention can be used to select an evaluation model for detecting the presence or absence of a specified abnormality based on process data obtained from various sensors installed in various locations in plant 1 to detect temperature, pressure, flow rate, valve opening, damper opening, liquid level, vibration, sound, and other state quantities of plant 1.

[0064] The evaluation results can be displayed in various formats. For example, the operating status of the plant may be scored according to the degree of deviation from the optimum region and displayed as an index called healthiness. The evaluation results may also be displayed as text information.

[0065] Furthermore, the display screen may display parameter values ​​(including hyperparameters) that can be set when performing machine learning based on training data, so that the operator can set them. Machine learning parameters vary depending on the machine learning model. For example, the display device 60 may be configured to display the number of layers in a decision tree, the number of units in a hidden layer in a neural network, parameters indicating weight decay, and the like, so that the operator can set them. For example, when the operator presses "Add New Model" in area AR7 (an example of a "seventh display area") on the display screen D1 using the input means, the display device 60 transitions to a display screen for setting new parameter values. When the operator inputs new parameter values, the model learning unit 44C performs machine learning according to the parameter values ​​and generates a new evaluation model. The operator can also set new training data along with the parameter values. For example, the operator can change the period of the training data on the display screen and set training data for a shorter or longer period. The model learning unit 44C performs machine learning based on the set new training data and generates a new evaluation model.

[0066] The display device 60 may be configured to allow the operator to directly set and input parameter values ​​or at least a portion of the learning data in area AR7 of the display screen D1 without screen transitions. For example, if overfitting of the evaluation model is observed in area AR3, the operator can change the parameter values ​​in area AR7 to mitigate the overfitting. Furthermore, the operator can set longer-term learning data in area AR7 to mitigate the overfitting. The model learning unit 44C performs machine learning according to the newly set parameter values ​​or learning period to generate a new evaluation model. Furthermore, the display device 60 can display the evaluation results of the new evaluation model in area AR3. This configuration allows the operator to observe the evaluation results while adjusting the range of parameter values ​​or learning data, thereby efficiently obtaining an appropriate evaluation model.

[0067] Furthermore, the display device 60 may be configured to allow the operator to delete or copy the evaluation model selected by the operator. For example, by configuring the display device 60 to allow the operator to delete an evaluation model that cannot be evaluated accurately, it becomes possible to efficiently obtain an appropriate evaluation model. Also, by copying an existing evaluation model and evaluating it by changing the evaluation criteria of the copied evaluation model, it becomes possible to efficiently obtain an appropriate evaluation model.

[0068] FIG. 4 is a block diagram showing a physical configuration for implementing the evaluation system 30. The evaluation system 30 includes a central processing unit (CPU) 30A corresponding to a calculation unit, a random access memory (RAM) 30B corresponding to a storage unit, a read-only memory (ROM) 30C corresponding to a storage unit, a communication unit 30D, an input unit 30E, and a display unit 30F. These components are connected to each other via a bus or a communication network so that data can be transmitted and received. While the present example describes the evaluation system 30 being configured with a single computer, the evaluation system 30 may also be implemented by combining multiple computers. For example, in addition to the display unit 30F, a display constituting a different display unit for displaying other information may be provided. The configuration shown in FIG. 4 is merely an example, and the evaluation system 30 may include other components or may not include some of these components. Furthermore, some of the components may be provided in a remote location. For example, all or part of the storage unit 30C may be provided in a remote location. In this case, the CPU 30A may be configured to acquire information from the storage unit 30C provided at a remote location via a communication network.

[0069] The CPU 30A is a calculation unit equipped with a computer processor that executes a computer program stored in the RAM 30B or the ROM 30C and performs each calculation process described in this embodiment. That is, the CPU 30A realizes each function executed by each unit of the model selection unit 44 in accordance with the computer program, and allows an operator to obtain an appropriate evaluation model. The CPU 30A may also realize a monitoring function in which the monitoring device 50 monitors the plant 1 based on process data under the selected evaluation model. The CPU 30A may also realize each function executed by the DCS 20. The CPU 30A receives various data from the input unit 30E and the communication unit 30D, displays the calculation results of the data on the display unit 30F, and stores the results in the RAM 30B or the ROM 30C.

[0070] The RAM 30B is a storage element that rewritably records information such as data and may be composed of semiconductor storage elements such as DRAM, SRAM, MRAM, NOR memory, or NAND memory. The RAM 30B can realize the functions of the storage unit 42 by storing the computer program executed by the CPU 30A, process data of the plant 1, cleansing rules, trained models, etc. The primary storage element such as DRAM or SRAM may be configured to store evaluation results based on the evaluation model, etc. The ROM 30C is a storage element from which information can be read and may be composed of, for example, a semiconductor storage element such as a write-protected NOR memory, a magnetic storage medium such as a HDD, or an optical storage medium such as a DVD. The ROM 30C may store the computer program executed by the CPU 30A and other data that cannot be rewritten.

[0071] The communication unit 30D is an interface that connects the evaluation system 30 to the DCS 20 and other devices. The communication unit 30D may be connected to a communication network such as the Internet.

[0072] The input unit 30E receives data input from the operator, and may be configured with, for example, a keyboard, a touch panel, or a microphone.

[0073] The display unit 30F visually displays the results of various arithmetic processing performed by the CPU 30A in this embodiment and the monitoring status of the plant 1 performed by the monitoring device 50, and may be configured with a display such as an LCD (Liquid Crystal Display). The display unit 30F realizes each function executed by the display device 60.

[0074] The computer programs for executing the various processes described in this embodiment may be stored in a computer-readable storage medium such as the ROM 30C and provided, or may be provided via a communication network connected by the communication unit 30D. In the display system 30, the CPU 30a executes the monitoring program to realize the various operations described in this embodiment. Note that these physical configurations are merely examples and do not necessarily have to be independent configurations. For example, the CPU 30A and RAM 30B may be configured as a packaged LSI (Large-Scale Integration).

[0075] The evaluation method for the evaluation model according to this embodiment will be described below. Fig. 5 is a flowchart showing an example of the evaluation method for comparing an existing evaluation model with a newly generated evaluation model. First, the evaluation system 30 (CPU 30A operating according to a computer program) reads existing trained models stored in the trained model storage unit 42C and displays the existing trained models in area AR1 of the display device 60. The operator uses the input unit 30E to select one or more trained models (sometimes referred to as "evaluation models") from the multiple trained models displayed in area AR1 (step S41). Area AR81 or area AR82 displays information about the training data used to generate the selected trained model, allowing the operator to select an appropriate trained model. The model selection unit 44D accepts the operator's input and selects a trained model to be evaluated.

[0076] Next, the worker inputs period information using the input unit 30E and registers the period of learning data for generating a new trained model (sometimes referred to as the "learning period"). The learning period registration unit 44A accepts the worker's input, registers the learning period, and acquires process data corresponding to this learning period from the process data storage unit 42A (step S42). Note that the worker may input period information by permitting the worker to input information into area AR81 or area AR82. Note that the learning periods may be different for each trained model, may be the same, or may be set so as to overlap partially.

[0077] The cleansing unit 44B cleanses the process data corresponding to the learning period in accordance with the cleansing rules stored in the cleansing rule storage unit 42B, and removes data that is not suitable for machine learning and process data that is not used for machine learning (for example, process data that does not affect or has a small effect on a predetermined abnormality when generating an evaluation model for detecting the abnormality) (step S43).

[0078] The model learning unit 44C performs machine learning based on the training data cleansed by the cleansing unit 44B to generate a new trained model (step S44). At this time, the evaluation system 30 may be configured so that an operator can set parameter values ​​of the machine learning model (e.g., the number of layers of a neural network). The trained model generated by the model learning unit 44C is stored in the trained model storage unit 42C.

[0079] Next, the operator inputs thresholds into the areas AR4 and AR5 using the input unit 30E. The threshold registration unit 44F receives the operator's input and registers the thresholds for each trained model (step S45).

[0080] Next, the operator uses the input unit 30E to input the evaluation period for the plant 1 in the area AR6. The evaluation period registration unit 44D accepts the operator's input, registers the evaluation period, and acquires the process data acquired during the evaluation period from the process data storage unit 42A as evaluation data (step S46). For example, by selecting an evaluation period that includes the time when an abnormality occurred in the plant 1, the operator can evaluate whether the trained model to be evaluated can correctly detect the abnormality.

[0081] The model evaluation unit 44G evaluates the evaluation data based on an existing trained model. The evaluation result display unit 44H provides information contained in area AR2, including the evaluation results, to the display device 60. At the same time, the model evaluation unit 44G evaluates the evaluation data based on the new trained model generated in step S44 and the threshold set in step S45. The evaluation result display unit 44H provides information contained in area AR3, including the evaluation results, to the display device 60 (step S47). The display device 60 receives the evaluation results from the evaluation system 30 and displays the evaluation results in areas AR2 and AR3 (step S48).

[0082] Next, the operator looks at the evaluation results displayed in areas AR2 and AR3 and determines whether the newly generated trained model is appropriate for monitoring plant 1 (step S49). If necessary, the operator may change the evaluation period or threshold and re-run the evaluation to determine whether the trained model is appropriate. For example, if the evaluation results using the newly generated trained model more accurately evaluate the operating status of plant 1 than the evaluation results using the existing trained model, it is possible to determine that the newly generated trained model is appropriate.

[0083] If it is determined to be appropriate, the operator uses the input unit 30E to press the "Production Model" setting button displayed in area AR4 or area AR5 corresponding to the newly generated trained model. The trained model storage unit 42 of the evaluation system 30 stores the newly generated trained model (step S50). From the next time onward, the newly generated trained model can be displayed in area AR1 as one of the existing trained models. Furthermore, the monitoring device 50 may start monitoring the plant 1 based on the process data acquired from the DCS 20 according to one or more trained models, including the newly generated trained model.

[0084] If it is determined in step S49 that the generated trained model is inappropriate, the process returns to step S42. In addition to re-registering the learning period, a new trained model may be generated by changing the parameters of the machine learning model.

[0085] As described above, the display device, evaluation method, and evaluation system according to the present embodiment make it possible to efficiently acquire an appropriate evaluation model. In particular, it is possible to easily create a new evaluation model on the same screen without switching screens, and it is also possible to display evaluation results using a new evaluation model and evaluation results using an existing evaluation model for the same evaluation data on the same screen. Furthermore, by changing parameters, it is possible to adjust an existing evaluation model or a new evaluation model, easily generate a different evaluation model, and compare the evaluation results with evaluation results using other evaluation models. As such, the display device, evaluation method, and evaluation system according to this embodiment make it easy to create an evaluation model, adjust the evaluation model, and compare evaluation models, making it possible to efficiently obtain an appropriate evaluation model.

[0086] The present invention can be used for purposes other than selecting an evaluation model for detecting a plant abnormality known as a blowout. For example, the present invention can be applied to detecting a decrease in power generation efficiency, clogged pipes, temperature abnormalities in plant elements such as furnaces, changes in the viscosity of fluids such as gases, or to selecting an evaluation model for determining the possibility of detection. In addition to detecting abnormalities, the present invention can also be applied to selecting and generating an evaluation model for improving processes performed in facilities such as plants based on process data.

[0087] Furthermore, the present invention can be applied to processes carried out in facilities other than thermal power plants. For example, the present invention can be applied to the selection and generation of an evaluation model for evaluating, based on process data, abnormalities (e.g., signs of gas leakage due to rot in pipes) related to processes carried out in facilities such as chemical plants, oil refineries, refineries, and steel factories, or parts thereof. For example, the present invention can be applied to selecting and generating an evaluation model for evaluating the possibility of fluid leakage based on process data such as temperature information acquired at a predetermined sampling frequency using multiple types of sensors, such as multiple temperature sensors and AE sensors, arranged in a piping system in which branch pipes are welded to a main pipe.

[0088] Furthermore, various modifications of the present invention are possible without departing from the spirit and scope of the present invention. For example, some components or functions in a certain embodiment can be replaced with other components or functions within the scope of the ordinary creative ability of a person skilled in the art. For example, in this embodiment, the model training unit 44C generates a new trained model through machine learning, but this is not limited to this. For example, the model training unit 44C may be configured to generate a trained model based on a regression equation. Also, it is not necessary to execute each step shown in the flowchart. [Explanation of symbols]

[0089] 1. Plant 2 Furnace 3. Cyclone 4. Recycling material recovery pipe 5 Rear flue 6 Furnace wall tube 7. Pump 8 Steam Drums 10 Superheater 12 Economizer 30 Rating System 40 Evaluation model management device 42 Storage section 50 Monitoring equipment 60 Display device 100 turbine 102 Condenser

Claims

1. a model selection unit that displays a plurality of selectable evaluation models in a first display area of ​​a display screen of the display device and accepts selection of the evaluation model; an evaluation result display unit that inputs evaluation data into each of the selected first evaluation model and the selected second evaluation model and evaluates them, and displays a first evaluation result of the evaluation data using the selected first evaluation model in a second display area of ​​the display screen, and displays a second evaluation result of the evaluation data using the selected second evaluation model in a third display area of ​​the display screen; Equipped with The plurality of evaluation models are generated based on the same learning data over different learning periods. Control device.

2. The evaluation result display unit further displaying a first evaluation criterion for evaluation using the first evaluation model in a settable manner in a fourth display area of ​​the display screen, and displaying a second evaluation criterion for evaluation using the second evaluation model in a fifth display area of ​​the display screen; the second display area is configured to be able to display the first evaluation result of the evaluation data by the selected first evaluation model based on the set first evaluation criterion, and the third display area is configured to be able to display the second evaluation result of the evaluation data using the selected second evaluation model based on the set second evaluation criterion. The control device according to claim 1 .

3. The evaluation result display unit Furthermore, a period of the evaluation data is displayed in a settable manner in a sixth display area of ​​the display screen, the second display area is configured to be able to display the first evaluation result of the evaluation data for the set period according to the first evaluation model, and the third display area is configured to be able to display the second evaluation result of the evaluation data for the set period according to the second evaluation model. The control device according to claim 1 or 2.

4. the first evaluation model is generated based on a range of training data by a machine learning model having predetermined parameters; The control device and further configured to display the parameter values ​​and the learning data ranges in a configurable manner in a seventh display area of ​​the display screen, the second display area is configured to be able to display an evaluation result of the evaluation data using an evaluation model generated by the machine learning model having the set parameter values; The control device according to any one of claims 1 to 3.

5. the first display area is configured to selectably display the machine learning model having the set parameter values ​​in the first display area together with the plurality of evaluation models; The control device according to claim 4.

6. The first display area is configured to be able to display information indicating that the evaluation model has been used in the past to evaluate process data acquired from a control object, and information indicating that the evaluation model is currently being used to evaluate process data acquired from the control object. The control device according to any one of claims 1 to 5.

7. the first display area is configured to display an evaluation model selected from the plurality of evaluation models in a manner that allows the evaluation model to be duplicated or deleted; The control device according to any one of claims 1 to 6.

8. a step of displaying a plurality of selectable evaluation models in a first display area of ​​a display screen; displaying an evaluation result of the evaluation data using the selected first evaluation model in a second display area of ​​the display screen; and displaying an evaluation result of the evaluation data using the selected second evaluation model in a third display area of ​​the display screen; The plurality of evaluation models are generated based on the same learning data over different learning periods. Evaluation method.

9. An evaluation system comprising one or more computers, a storage means for storing a plurality of evaluation models; a selection means for displaying a selectable evaluation model to be used for evaluation from the plurality of evaluation models and accepting selection of the evaluation model; evaluation means for inputting evaluation data into the selected first evaluation model to obtain a first evaluation result; evaluation means for inputting the evaluation data into the selected second evaluation model to obtain a second evaluation result; a display means for displaying the first evaluation result and the second evaluation result; Equipped with The plurality of evaluation models are generated based on the same learning data over different learning periods. Rating system.

Images