WO2021140742A1 - Dispositif de support de gestion de fonctionnement et procédé de support de gestion de fonctionnement - Google Patents

Dispositif de support de gestion de fonctionnement et procédé de support de gestion de fonctionnement Download PDF

Info

Publication number
WO2021140742A1
WO2021140742A1 PCT/JP2020/042333 JP2020042333W WO2021140742A1 WO 2021140742 A1 WO2021140742 A1 WO 2021140742A1 JP 2020042333 W JP2020042333 W JP 2020042333W WO 2021140742 A1 WO2021140742 A1 WO 2021140742A1
Authority
WO
WIPO (PCT)
Prior art keywords
control value
unit
actual
value
reliability
Prior art date
Application number
PCT/JP2020/042333
Other languages
English (en)
Japanese (ja)
Inventor
亜由美 渡部
晃治 陰山
豊 三宮
原　直樹
信幸 中村
Original Assignee
株式会社日立製作所
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 株式会社日立製作所 filed Critical 株式会社日立製作所
Publication of WO2021140742A1 publication Critical patent/WO2021140742A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B23/00—Testing or monitoring of control systems or parts thereof
    • G05B23/02—Electric testing or monitoring

Definitions

  • the present invention relates to a technique of an operation management support device and an operation management support method.
  • Operation management of plants, etc. that set control values based on measured values is being performed.
  • operation management generally, a method of deriving a predicted control value by using a simulation model simulating a physical phenomenon or a method of deriving a predicted control value by utilizing statistics or machine learning based on actual data is used. Be done.
  • Patent Document 1 presents a model identification device, a prediction device, a monitoring system, a model identification method, and a prediction method.
  • the model identification device adds a label of the mode to which it belongs to a part of a plurality of input / output data, classifies the input / output data into each mode (clustering), estimates the parameters of the mathematical model of each mode, and clusters. Based on the results of the above, there is disclosed an apparatus for model identification by classifying input data into each of the mathematical models.
  • the reliability of the derived predicted control values greatly depends on the characteristics of the actual data used. For example, when the measured values when the plant is operated under different conditions from when the actual data was acquired are input to the model obtained by learning, the reliability of the predicted control values by the model created based on the actual data. Will be low. However, when the predictive control value is calculated by utilizing statistics and machine learning, it is difficult to grasp its reliability because the predictive control value is calculated as a numerical value. The technique described in Patent Document 1 cannot solve such a problem.
  • the present invention has been made in view of such a background, and an object of the present invention is to perform efficient operation management of a device.
  • the present invention presents the actual measurement value which is the measured value of the device in the past and the actual result which is the control value of the device in the past among the past actual data stored in the storage unit.
  • the first model generation unit that generates a first model for calculating the prediction control value using the control value and the first model generated by the first model generation unit.
  • the predictive control value calculation unit that calculates the predictive control value corresponding to the measured value by applying the input measured value and the actual data stored in the storage unit, the actual measured value is used.
  • the second model generation unit that generates the second model and the category corresponding to the measured value input from the device are classified by the second model generation unit.
  • the category selection unit selected based on the category, the prediction control value calculated by the prediction control value calculation unit, and the actual control value in the category selected by the category selection unit are compared. It is characterized by including a reliability calculation unit for calculating the reliability of the prediction control value and a display processing unit for displaying the reliability calculated by the reliability calculation unit on the display unit.
  • a reliability calculation unit for calculating the reliability of the prediction control value and a display processing unit for displaying the reliability calculated by the reliability calculation unit on the display unit.
  • the embodiment a mode for carrying out the present invention (referred to as "the embodiment") will be described in detail with reference to the drawings as appropriate.
  • the same or similar configurations are designated by the same reference numerals, and if the explanations are duplicated, the explanations may be omitted. Further, also for each flowchart, the same step number may be assigned to the same processing, and if the description is duplicated, the description may be omitted.
  • a model is generated by these learnings.
  • the input of the actual control value as the objective variable is indispensable for model creation is defined as supervised learning.
  • unsupervised learning is to create a model using only the input items (explanatory variables) selected from the actual measurement values. The actual control value and the actual measurement value will be described later.
  • the application is not limited to the target.
  • the operation management support device 1 of the present embodiment can be applied in addition to the plant PL such as a water treatment system.
  • FIG. 1 is a diagram showing a configuration example of the operation management support device 1 according to the first embodiment.
  • the operation management support device 1 has a teachered learning processing unit 110 and a teacherless learning processing unit 120 that performs unsupervised learning.
  • the operation management support device 1 has a reliability evaluation processing unit 130 that calculates a reliability 232 based on the result of the supervised learning processing unit 110 and the result of the unsupervised learning processing unit 120.
  • the operation management support device 1 has a display processing unit 141 that displays the reliability 232 output from the reliability evaluation processing unit 130 on the monitor 142.
  • the supervised learning processing unit 110 performs supervised learning on the performance control data 202 and the performance measurement data 203 among the performance data 201 stored in the performance DB 101.
  • the supervised learning processing unit 110 has a prediction model construction unit 111 and a prediction control value calculation unit 112. The processing performed by the prediction model construction unit 111 and the prediction control value calculation unit 112 will be described later.
  • the actual control data 202 is data including a plurality of past actual control values
  • the actual measurement data 203 is data including a plurality of past actual measurement values.
  • the measured values are, for example, the temperature at the point A, the flow rate at the point B, the flow rate at the point C, the concentration of the component E in the raw material D, the input amount of the raw material D, and the like in the plant PL.
  • the operation pattern may be left in the measured value as flag information.
  • the operation pattern is, for example, at the start of operation, at steady operation, at stop of operation, and the like.
  • the control value is, for example, the valve opening degree of a certain valve in the plant PL, the input amount setting value of the raw material C, and the like. In the present embodiment, the valve opening degree is used as an example of the control value.
  • the actual control value is a past control value collected in the actual DB 101
  • the actual measurement value is a past measurement value collected in the actual DB 101.
  • the unsupervised learning processing unit 120 performs unsupervised learning on the performance measurement data 203 of the performance data 201 stored in the performance DB 101.
  • the unsupervised learning processing unit 120 has a classification model construction unit 121, a classification result calculation unit 122, and a category information generation unit 123. The processing performed by the classification model construction unit 121, the classification result calculation unit 122, and the category information generation unit 123 will be described later.
  • the reliability evaluation processing unit 130 has a deviation degree calculation unit 131 and a reliability calculation unit 132. The processing performed by the deviation degree calculation unit 131 and the reliability calculation unit 132 will be described later.
  • the operation management support device 1 of the first embodiment derives the prediction control value y1 by the prediction model 211 by supervised learning based on the actual data 201. At the same time, the operation management support device 1 calculates the category information y2 by unsupervised learning based on the actual measurement data 203, and calculates the reliability 232 by comparing with the prediction control value y1.
  • the measurement data 204, the classification model 221 and the search result 222, and the deviation degree 231 will be described later.
  • FIG. 2 is a flowchart showing a procedure of processing performed by the operation management support device 1 shown in FIG. First, the past measured values (actual measured values) and control values (actual control values) of the plant PL collected in advance are stored in the actual DB 101 as actual data 201.
  • the prediction model construction unit 111 acquires a part or all of the actual data 201 from the actual DB 101 (S101).
  • the actual data 201 includes a part or all of the actual measurement data 203 and a part or all of the actual control data 202.
  • the prediction model construction unit 111 performs supervised learning on the input performance measurement data 203 and the performance control data 202, generates the prediction model 211 (S102), and outputs the prediction model 211.
  • the prediction model 211 generated by the prediction model construction unit 111 is input to the prediction control value calculation unit 112, and the measurement data 204 is input from the plant PL. Then, the prediction control value calculation unit 112 calculates and outputs the prediction control value y1 by substituting the measurement data 204 into the prediction model 211 (S103).
  • the measurement data 204 includes the measurement values measured by the plant PL.
  • supervised learning for example, a regression equation, a decision tree, a random forest, a convolutional neural network, a neural network including a recurrent neural network, and the like can be used.
  • the actual measurement values of the actual measurement data 203 may be used as they are, or the items selected from the values of the actual measurement data 203 under predetermined conditions may be used.
  • the difference or total value of the two items, the integrated value or the average value of one item for a certain period may be input to the supervised learning processing unit 110 as an input item.
  • the item is a type of measured value such as temperature and flow rate.
  • the prediction model 211 generated by the prediction model construction unit 111 means the structure of the generated model and each variable in the optimized model.
  • the classification model construction unit 121 acquires the actual measurement data 203 stored in the actual DB 101 (S111). Then, the classification model construction unit 121 generates (S112) a classification model 221 (category) by performing unsupervised learning on the acquired performance measurement data 203, and outputs the classification model 221 (S112).
  • the classification model 221 classifies the actual measurement data 203 into categories according to predetermined conditions. As the category, the previously generated category may be used.
  • the classification result calculation unit 122 searches for the category to which the measurement data 204 acquired from the plant PL belongs (S113).
  • the classification result calculation unit 122 outputs the search result 222 to the category information generation unit 123.
  • the search result 222 includes information on the category to which the measurement data 204 belongs.
  • the measurement data 204 includes the measured values measured by the plant PL.
  • the category information generation unit 123 generates and outputs the category information y2 based on the search result 222 (S114).
  • the category information y2 output here includes a plurality of values such as the maximum value and the minimum value of the category to which the measurement data 204 belongs, and all the actual control values included in the category.
  • unsupervised learning for example, hierarchical clustering, non-hierarchical clustering including the K-means method, a neural network including a self-organizing map, and the like can be used.
  • the performance measurement data 203 to be input without teacher learning the past performance measurement values as described above may be used as input items as they are, or the performance measurement values of the selected predetermined items may be used.
  • the difference or total value of the two items in the actual measurement value, the integrated value or the average value of one item for a certain period may be input instead of the actual measurement data 203.
  • the deviation degree calculation unit 131 acquires the prediction control value y1 from the prediction control value calculation unit 112 and the category information y2 from the category information generation unit 123. Then, the deviation degree calculation unit 131 calculates the deviation degree 231 between the acquired prediction control value y1 and the category information y2 (S121).
  • any of the following equations (1) to (3) is used to calculate the degree of divergence 231.
  • at least two of the formulas (1) to (3) may be combined.
  • an expression other than the equations (1) to (3) may be used as the calculation expression of the deviation degree 231.
  • D y2min-y1 ... (1)
  • D y1-y2max ... (2)
  • D
  • D is the degree of divergence 231.
  • y1 is a prediction control value output from the prediction control value calculation unit 112.
  • y2ave is an average value of category information y2.
  • y2min is the minimum value of the category information y2.
  • y2max is the maximum value of the category information y2.
  • the category information y2 is generally output as a value having a predetermined width.
  • the reliability calculation unit 132 calculates the reliability 232 based on the deviation degree 231 output from the deviation degree calculation unit 131 (S122) and outputs it.
  • the reliability 232 is calculated so that the larger the deviation degree 231 is, the lower the reliability is, and the smaller the deviation degree 231 is, the higher the reliability degree is.
  • the reliability calculation unit 132 calculates the reliability 232 by, for example, setting an arbitrary conversion formula or a predetermined reference value and using a correspondence table of the reliability 232 with respect to the deviation degree 231 in an arbitrary range. For example, when the equation (1) is used as the deviation degree 231, the following equation (11) is used as the conversion equation of the reliability 232.
  • the degree of divergence 231 is large, the degree of divergence is low, and if the degree of divergence 231 is small, the degree of divergence is high.
  • the display processing unit 141 displays the calculated reliability 232 on the monitor 142 together with the prediction control value y1 and the category information y2 (S131). As a result, the operator can determine the control value to be set in the plant PL by referring to the reliability 232 as well as the predicted control value y1.
  • the predictive control value y1 is calculated by supervised learning, but the predictive control value y1 may be calculated by unsupervised learning. That is, the supervised learning processing unit 110 may be replaced with one that outputs the predictive control value y1 by unsupervised learning.
  • unsupervised learning hierarchical clustering, non-hierarchical clustering including the K-means method, a neural network including a self-organizing map, and the like are used.
  • the actual control data 202 is classified into categories, and it is searched which category of the actual control data 202 the control value acquired from the plant PL belongs to.
  • the classification model is the prediction model. Then, the average value, the median value, the minimum value, the maximum value, and the like of the actual control values in the category to which the acquired control value belongs are output as the predicted control value y1. This also applies to the following second embodiment and below.
  • the operation management support device 1 of the first embodiment based on the comparison between the predicted control value y1 calculated by the supervised learning processing unit 110 and the category information y2 calculated by the unsupervised learning processing unit 120, The reliability 232 of the prediction control value y1 is calculated. Then, the operator can determine whether or not the predictive control value y1 can be applied from the reliability 232 of the predictive control value y1. Therefore, it is possible to prevent a decrease in operating efficiency, a decrease in product quality, a defect, and the like due to control using the unreliable predicted control value y1.
  • the driving management support device 1a in which three or more models including supervised learning and unsupervised learning are used will be described.
  • FIG. 3 is a diagram showing a configuration example of the operation management support device 1a according to the second embodiment.
  • the differences between the operation management support device 1a and the operation management support device 1 shown in FIG. 1 are as follows.
  • (A1) Two supervised learning processing units 110 are provided (first supervised learning processing unit 110A and second supervised learning processing unit 110B).
  • the first supervised learning processing unit 110A has a first prediction model construction unit 111A and a first prediction control value calculation unit 112A.
  • the second supervised learning processing unit 110B has a second prediction model construction unit 111B and a second prediction control value calculation unit 112B.
  • the first unsupervised learning processing unit 120A has a first classification model construction unit 121A, a first classification result calculation unit 122A, and a first category information generation unit 123A.
  • the second unsupervised learning processing unit 120B has a second classification model construction unit 121B, a second classification result calculation unit 122B, and a second category information generation unit 123B.
  • A3 Determination of a supervised learning representative value for calculating the representative value x1 of the first predictive control value y1A and the second predictive control value y1B output from the first supervised learning processing unit 110A and the second supervised learning processing unit 110B.
  • a section 151 is provided. Further, the teacher non-learning representative value determination unit 152 that calculates the representative information x2 of the first category information y2A and the second category information y2B output from the first teacher non-learning processing unit 120A and the second teacher non-learning processing unit 120B It is provided.
  • the supervised learning used in the first supervised learning processing unit 110A and the second supervised learning processing unit 110B is a regression equation, a decision tree or a random forest, a convolutional neural network, or a recurrent neural network, as in the first embodiment.
  • a neural network or the like including is used.
  • the same type of learning may be used for the supervised learning used in the first supervised learning processing unit 110A and the second supervised learning processing unit 110B, or another type of learning may be used.
  • the input items input to the supervised learning may be the same or different items.
  • the unsupervised teachers used in the first unsupervised learning processing unit 120A and the second unsupervised learning processing unit 120B are hierarchical clustering, non-hierarchical clustering including the K-means method, and self-organizing as in the first embodiment.
  • a neural network or the like including a clustering map is used.
  • the same type of learning may be used for the unsupervised learning used in the first teacher non-learning processing unit 120A and the second teacher non-learning processing unit 120B, or another type of learning may be used.
  • the input items to be input without teacher learning may be the same or different items.
  • FIG. 4 is a flowchart showing a procedure of processing performed by the operation management support device 1a shown in FIG. (Processing of the first supervised learning processing unit 110A and the second supervised learning processing unit 110B)
  • the first supervised learning processing unit 110A outputs the first predictive control value y1A by performing the same processing as in steps S101 to S103 of FIG.
  • the second prediction control value y1B is output by performing the same processing as in steps S101 to S103 of FIG.
  • the supervised learning representative value determination unit 151 acquires the first predictive control value y1A and the second predictive control value y1B. Subsequently, the supervised learning representative value determination unit 151 calculates (S201) a representative value x1 of the first predictive control value y1A and the second predictive control value y1B, and outputs the result.
  • the representative value x1 is any one of an average value, a median value, a maximum value, and a minimum value. Further, the representative value x1 is treated as the prediction control value y1.
  • the first category information y2A is output by performing the same processing as in steps S111 to S114 of FIG.
  • the second category information y2B is output by performing the same processing as in steps S111 to S114 of FIG.
  • the unsupervised learning representative value determination unit 152 acquires the first category information y2A and the second category information y2B. Subsequently, the unsupervised learning representative value determination unit 152 calculates (S202) and outputs the representative information x2 of the first category information y2A and the second category information y2B.
  • the representative information x2 is, for example, a set of the larger of the maximum values of the first category information y2A and the second category information y2B and the smaller of the minimum values of the first category information y2A and the second category information y2B. And so on.
  • the average value of the first category information y2A and the second category information y2B, or a set of medians may be output as representative information x2.
  • the wider standard deviation or the narrower standard deviation may be output as the representative information x2.
  • the representative information x2 is treated as the category information y2.
  • the deviation degree calculation unit 131 acquires the representative value x1 and the representative information x2, and calculates the deviation degree 231 based on the representative value x1 and the representative information x2 (S121a).
  • the representative value x1 is used instead of the prediction control value y1 and the representative information x2 is used instead of the category information y2 in the process of step S121 of FIG. Since step S122 and subsequent steps are the same as the process of FIG. 2, the description here will be omitted.
  • more accurate predictive control value y1 (representative value x1) and reliability 232 can be obtained by performing a plurality of supervised learning and / or a plurality of unsupervised learning.
  • two teachered learning processing units 110 and two unsupervised learning processing units 120 are provided, but three or more teachered learning processing units 110 may be provided. Further, three or more teacher non-learning processing units 120 may be provided. Alternatively, one of the supervised learning processing unit 110 and the unsupervised learning processing unit 120 may be provided. Also in these forms, the reliability 232 can be calculated by the same procedure as that shown in FIG.
  • FIG. 5 is a diagram showing a configuration example of the operation management support device 1b according to the third embodiment.
  • the operation management support device 1b differs from the operation management support device 1 shown in FIG. 1 in that a reliability determination unit 161 and a setting processing unit 143 are provided after the reliability evaluation processing unit 130. The processing performed by the reliability determination unit 161 and the setting processing unit 143 will be described later.
  • FIG. 6 is a flowchart showing a procedure of processing performed by the operation management support device 1b shown in FIG. As shown in FIG. 6, after step S122, the reliability determination unit 161 performs the reliability determination process (S301). The reliability determination process will be described later.
  • FIG. 7 is a flowchart showing a detailed procedure of the reliability determination process in step S301 of FIG.
  • the reliability determination unit 161 acquires the reliability 232 output from the reliability calculation unit 132 (S311).
  • the reliability determination unit 161 classifies the acquired reliability 232 into the reliability 232 by the threshold value determination (S312).
  • the reliability 232 is classified into "low / medium / high", "applicability / non-applicability / examination required / applicable”, etc. according to a preset threshold value.
  • step S312 when the reliability 232 is "low” and “medium” (S312 ⁇ “low” and “medium”), the display processing unit 141 displays a warning on the monitor 142 (S321). After that, the display processing unit 141 displays the reliability 232, the prediction control value y1, and the category information y2 on the monitor 142 (S322).
  • step S312 when the reliability 232 is "high" as a result of step S312 (S312 ⁇ high), the display processing unit 141 displays the reliability 232, the prediction control value y1, and the category information y2 on the monitor 142 (). S331). Then, the setting processing unit 143 sets the predicted control value y1 as the control value of the plant PL (S332). The process of step S332 can be omitted.
  • step S322 and step S331 the threshold value used in step S312 may be displayed together with the reliability 232, the prediction control value y1, and the category information y2.
  • the reliability determination unit 161 ranks the reliability 232 by determining the threshold value of the reliability 232. By doing so, it is possible to obtain a standard as to whether or not the operator adopts the predicted control value y1 as the control value set in the plant PL. If it is determined that the reliability 232 is "high" as a result of the threshold value determination, the setting processing unit 143 sets the predicted control value y1 as the control value in the plant PL. By doing so, the application of the predicted control value y1 can be automated, and the burden on the operator can be reduced.
  • FIG. 8 is a diagram showing a configuration example of the operation management support device 1c according to the fourth embodiment.
  • the operation management support device 1c of the fourth embodiment is provided with a trend determination unit 170 after the reliability evaluation processing unit 130.
  • the trend determination unit 170 includes a control range determination unit 171 and a trend match determination unit 172.
  • a setting processing unit 143 is provided. The processing performed by the control range determination unit 171, the trend match determination unit 172, and the setting processing unit 143 will be described later.
  • FIG. 9 is a flowchart showing a procedure of processing performed by the operation management support device 1c shown in FIG. As shown in FIG. 9, after step S122, the trend determination unit 170 performs the trend determination process (S401). The trend determination process will be described later.
  • (Trend judgment processing) 10A and 10B are flowcharts showing a detailed procedure of the trend determination process in step S401 of FIG.
  • the control range determination unit 171 uses the actual control data 202 at the past N time
  • the trend match determination unit 172 uses the actual control data 202 at the past M time. At this time, N and M do not necessarily have to match.
  • the past time N used by the control range determination unit 171 is designated via the input device (see FIG. 15) 411 (S411 in FIG. 10A).
  • N is determined by the operator by referring to the time constant, empirical knowledge, etc. of the target plant PL.
  • the control range determination unit 171 acquires the actual control data 202 for the past N times from the actual DB 101 (S412).
  • control range determination unit 171 acquires the predicted control value y1 for the past N times (S413). Next, the control range determination unit 171 determines whether or not all of the acquired predicted control values y1 are included in the control range (S414). For example, the plant PL is controlled within ⁇ y (or within ⁇ y%) of the change amount of the past valve opening degree (actual control value) at an arbitrary n time (n ⁇ N). In this case, ⁇ y becomes the control range, and in step S413, the control range determination unit 171 determines whether or not the predicted control value y1 is within the range of ⁇ y. Normally, such a control range is defined as a rule for the control value of the plant PL.
  • control range may be set separately, or the existing rule may be diverted as the control range. Further, when the predicted control value y1 is out of the control range (S414 ⁇ No), the display processing unit 141 displays on the monitor 142 that the control range determination result is an error (S415). Then, the operation management support device 1c ends the process.
  • the display processing unit 141 notifies the monitor 142 that the control range determination result is “Yes”. Display (S416). Then, the past time M used by the trend matching determination unit 172 is input (designated) via the input device (see FIG. 15) 411 (S421 in FIG. 10B). Next, the trend match determination unit 172 acquires the actual control data 202 of the past M time from the actual DB 101 (S422). Further, the trend matching determination unit 172 acquires the latest prediction control value y1 (S423).
  • the trend match determination unit 172 acquires the current trend from the actual control data 202 of the past M time acquired in step S422 (S424).
  • the current trend is extracted, for example, by calculating the difference or slope of the period from the present to the past time M in the actual control data 202, and then determining the decrease / maintenance / increase based on the threshold value.
  • the difference between the actual control values before m1 time and m2 time the difference between the actual control value before m1 time and the average value of the actual control values at the past M time, and the like can be used.
  • the current trend is extracted by the slope, the value calculated by changing the actual control value from m1 time to m2 time as the slope with respect to the time can be used.
  • m1 and m2 are m1 ⁇ m2 ⁇ M.
  • the trend match determination unit 172 calculates the difference or slope from the actual control data 202 at the past M-1 time and the latest predicted control value y1, and extracts the predicted trend of the predicted control value y1 based on the threshold value determination. (S425).
  • the forecast trend is obtained in the same way as the current trend.
  • the trend matching determination unit 172 determines whether or not the predicted trend matches the current trend (S431). Specifically, the trend matching determination unit 172 determines whether or not the matching rate between the predicted trend and the current trend is equal to or greater than a predetermined value. If the matching rate between the predicted trend and the current trend is equal to or greater than a predetermined value, the trend matching determination unit 172 determines in step S431 that the matching rate is “matching”. If the matching rate between the predicted trend and the current trend is less than a predetermined value, the trend matching determination unit 172 determines in step S431 that there is a “mismatch”.
  • the display processing unit 141 displays “trend match” on the monitor 142 as the trend determination result (S432).
  • the display processing unit 141 displays a warning on the monitor 142 (S433).
  • step S432 and step S433 the setting processing unit 143 determines whether or not the setting condition is satisfied (S441).
  • the setting condition is that it is determined as "OK” in step S414 and "matched” in step S431.
  • the setting processing unit 143 sets the latest predicted control value y1 as the control value in the plant PL (S442). If the setting condition is not satisfied (S441 ⁇ No), the operation management support device 1c returns the process to step S131 in FIG.
  • steps S415, S416, S432, and S433 the control range determination result and the trend determination result are output to the monitor 142. Further, the processes of steps S441 and S442 can be omitted. Further, the trend matching determination unit 172 may be omitted. In this case, if it is determined as “OK” in step S414, the setting processing unit 143 may set the latest predicted control value y1 in the plant PL as a control value. Alternatively, the control range determination unit 171 may be omitted. In this case, if it is determined as "match” in step S431, the setting processing unit 143 may set the latest predicted control value y1 in the plant PL as a control value.
  • the determination is executed in the order of the control range determination (S414) and the trend match determination (S431), but the determination may be executed in the order of the trend match determination and the control range determination.
  • the operator can determine whether or not to adopt the predicted control value y1 in consideration of the control range of the current operating status and the trend. Further, based on the control range and the trend, it is determined whether or not the predicted control value y1 is set as the control value of the plant PL. Then, if the predicted control value y1 is within the control range or the current trend and the predicted trend match, the setting processing unit 143 sets the predicted control value y1 as the control value of the plant PL. As a result, the work load of the operator can be reduced.
  • FIG. 11 is a diagram showing a configuration example of the operation management support device 1d according to the fifth embodiment.
  • the operation management support device 1d is different from the operation management support device 1 shown in FIG. 1 in the following points.
  • C1 A data feature determination unit 181 for inputting actual data 201 from the actual DB 101 is provided.
  • C2 A data selection / classification unit 182 and a deviation degree calculation method selection unit 183 are provided downstream of the data feature determination unit 181.
  • C3 A supervised learning processing unit 110d having a prediction model construction unit 111d and a prediction control value calculation unit 112d is provided.
  • a teacher non-learning processing unit 120d having a classification model construction unit 121d, a classification result calculation unit 122d, and a category information generation unit 123 is provided.
  • the processing performed by the prediction model construction unit 111d, the prediction control value calculation unit 112d, the classification model construction unit 121d, the classification result calculation unit 122d, the data feature determination unit 181 and the data selection classification unit 182, and the deviation degree calculation method selection unit 183. Will be described later.
  • FIG. 12 is a flowchart showing a procedure of processing performed by the operation management support device 1d shown in FIG.
  • the data feature determination unit 181 first extracts the features of the actual data 201 (S501).
  • the feature of the extracted actual data 201 is statistical information such as the average value, variance, and distribution shape of all the actual data 201 or some of the actual data 201 stored in the actual DB 101.
  • the data feature determination unit 181 may target some actual data 201 such as a steady operation period of the plant PL as a target for feature extraction based on the flag information of the operation pattern and the operation period.
  • the statistical information of a part of the target actual data 201 may be a feature of the actual data 201.
  • the difference between the statistical information of some of the actual data 201 and the statistical information of all the actual data 201 may be a feature of the actual data 201.
  • the data selection classification unit 182 selects and classifies the actual data 201 (S502), and sends a group of the actual data 201, which is the result of the selection and classification, to the supervised learning processing unit 110 and the unsupervised learning processing unit 120.
  • the sorting classification by the data sorting classification unit 182 in step S502 for example, the following can be considered.
  • the conditions under which the sorting and classification are performed are set by the operator.
  • (C12) Sorting classification based on flag information For example, the data selection / classification unit 182 sorts and classifies the actual data 201 based on the flag information from the start of operation of the plant PL to 2 hours, 2 hours to 5 hours, 5 hours, and the end of operation of the plant PL.
  • the actual data 201 includes various types of data such as temperature, pressure (actual measurement value), valve opening degree, raw material input amount setting value (actual control value), and the like.
  • the selection and classification in step S502 can also be performed for each of these types of data.
  • the divergence degree calculation method selection unit 183 is based on the correspondence table 500 between the characteristics of the actual data 201 and the divergence degree calculation method, which are determined in advance, and the characteristics of the actual data 201 calculated by the data feature determination unit 181.
  • the method for calculating the degree of divergence is determined (S511).
  • the deviation degree calculation method selection unit 183 may adjust the parameters for calculating the deviation degree based on the characteristics of the actual data 201 calculated by the data feature determination unit 181.
  • Steps S101 to S103, S111 to S114, S121 to S122, and S131 are substantially the same as those in the first embodiment, but are different in the following points.
  • the prediction model construction unit 111d generates the prediction model 211 for each group of the actual data 201 divided by the selection classification in step S502 (S102a).
  • the predictive control value calculation unit 112d calculates and outputs the predictive control value y1 (S103a).
  • the group corresponding to the acquired measurement data 204 is, for example, as follows. When a group is formed based on the average value, it is a group corresponding to the average value of the measurement data 204. When a group is formed based on the flag information, the group corresponding to the flag information of the measurement data 204 becomes the group corresponding to the measurement data 204.
  • the measurement data 204 also includes various types of data such as temperature and pressure, but when the selection classification in step S502 is performed for each type of data, the calculated predictive control value y1 is the data. The selection classification for each type will be reflected. That is, the predictive control value calculation unit 112d calculates the predictive control value y1 based on the temperature group with respect to the temperature measurement data 204.
  • the classification model construction unit 121d generates a classification model 221 (category) for each group of actual data 201 divided by the selection classification in step S502 (S112a).
  • the classification result calculation unit 122d searches for a category. At this time, the category of the group corresponding to the value of the acquired measurement data 204 is searched (S113a). The group corresponding to the value of the measurement data 204 is the same as that described above in (C22). Further, when the selection classification in step S502 is performed for each data type, the category search reflects the data type. For example, the classification result calculation unit 122d searches the temperature measurement data 204 for a category in the temperature group.
  • FIG. 13 is a diagram showing an example of a correspondence table 500 between the features of the actual data 201 and the deviation degree calculation method.
  • the correspondence table 500 between the characteristics of the actual data 201 and the deviation degree calculation method shows the “data” having the “judgment target” and the “comparison target”, the “data characteristics” of the actual data 201, and the “deviation”. It has each item of "degree calculation method".
  • the deviation degree calculation method of the "A method” is applied when the statistical distribution is unevenly distributed on the negative side as a data feature in all the data.
  • the deviation degree calculation method of the B method is applied to all the data of the data flagged with "operation pattern A”. Has been done.
  • the prediction model 211 and the classification model 221 are generated based on the actual data 201 that is sorted and classified under predetermined conditions and divided into groups. Then, the prediction control value y1 and the category information y2 are calculated based on these prediction models 211 and the classification model 221. As a result, the calculation of the prediction control value y1 and the category information y2 reflects the group according to the selection classification in step S502. That is, the prediction control value y1 and the category information y2 are calculated based on the group of the actual data 201 having the same properties as the measurement data 204. Therefore, the accuracy of the reliability 232 calculated based on the prediction control value y1 and the category information y2 can be improved.
  • steps S101a, S102a, S111a, and S112a of FIG. 12 are the same processes as steps S101, S102, S111, and S112 of the first embodiment.
  • FIG. 14 is a diagram showing an example of the operation management screen 300 in the present embodiment.
  • the operation management screen 300 has a control target display area 311 and a predictive control value display area 320. Further, the operation management screen 300 has a category name display area 331, a reliability display area 332, a category information display area 333, and a control value frequency information display area 334. Further, the operation management screen 300 displays a reliability determination result display area 341, a trend determination result display area 342, a data feature display area 343, a deviation degree calculation method display area 344, a learning type display area 345 used, and a control state display area 351. Have.
  • control target display area 311 the equipment in the plant PL to be controlled, the parts in the equipment, and the control value are displayed.
  • the device to be controlled is the "valve A" of the "heat exchanger A”
  • the "opening" of the "valve A” of the “heat exchanger A” is the control target. It is shown.
  • the predicted control value display area 320 a graph of the actual control value and the predicted control value y1 is shown.
  • the actual control value is shown by a solid line graph
  • the predicted control value y1 is shown by a broken line graph.
  • the asterisk indicates the latest predicted control value y1.
  • the predictive control value display area 320 includes the latest predictive control value display area 321.
  • the value displayed in the latest predicted control value display area 321 is the predicted control value y1 corresponding to the star mark in the predicted control value display area 320.
  • the predicted control value y1 is also displayed as a percentage.
  • the category to which the measured value acquired from the plant PL belongs is displayed, which is classified without teacher learning.
  • FIG. 14 it is shown that the “opening” of the target “valve A” belongs to the “category A”.
  • the reliability 232 calculated by the reliability calculation unit 132 is displayed.
  • Information about each category generated as a result of unsupervised learning is displayed in the category information display area 333.
  • the valve opening degree which is a control value
  • the opening degree of the "valve A" to be controlled is indicated by a white circle, it is possible to visually recognize which category the control value to be controlled belongs to.
  • the control value frequency information display area 334 the frequency of the actual control value of the opening degree of all the valves in the plant PL is shown by a histogram. Then, the histogram corresponding to the calculated prediction control value y1 is shown by diagonal lines.
  • the determination result in step S301 of FIG. 6 is shown in the reliability determination result display area 341.
  • the threshold value used in step S301 of FIG. 6 may be displayed together with the determination result.
  • the trend determination result display area 342 shows the determination result in step S431 of FIG. 10B.
  • the determination result of step S414 of FIG. 10A may be displayed together with the determination result of step S431.
  • the name of the data feature determined by the data feature determination unit 181 of FIG. 11 is shown.
  • the name of the divergence degree calculation method determined by the divergence degree calculation method selection unit 183 of FIG. 11 is displayed.
  • the learning type display area 345 used the name of the learning used for deriving the prediction control value y1 or the category information y2 is displayed.
  • control status display area 351 information about what kind of control is being performed in the plant PL to be controlled is displayed.
  • the control state display area 351 is displayed to indicate that the automatic operation mode (automatic control) in which the predicted control value y1 is set as the control value is performed, but the predicted control value y1 is monitored. It is also possible to display the manual operation mode in which the operator confirms with 142 and sets the predicted control value y1 as the control value in the plant PL. Further, in the control state display area 351, the automatic operation mode and the manual operation mode may be switched by the information input by the operator via the input device (see FIG. 15) 411. The automatic operation mode can be executed in the third embodiment and the fourth embodiment.
  • the judgment result by the reliability judgment unit 161, the control range judgment unit 171 and the trend match judgment unit 172, and the coping method when the judgment result is other than "OK" are the control states. It may be displayed in the display area 351.
  • FIG. 15 is a diagram showing the hardware configurations of the operation management support devices 1, 1a to 1d.
  • the operation management support device 1 includes a memory 401, a CPU (Central Processing Unit) 402, a storage device 403, an input device 411, a communication device 412, and a monitor 142.
  • the communication device 412 transmits / receives information to / from the plant PL.
  • the storage device 403 includes the actual result DB 101, and each part 110 to 112, 110A to 112A, 110B to 112B, 120 to 123, 120A to 123A, 120B to 123B, 130 to 132, 141, 143, 151, 152, 161. , 170 to 172, 181 to 183 are stored.
  • the program stored in the storage device 403 is loaded into the memory 401. Then, the program loaded in the memory 401 is executed by the CPU 402 to execute each part 110 to 112, 110A to 112A, 110B to 112B, 120 to 123, 120A to 123A, 120B to 123B, 130 to 132, 141. 142, 151, 152, 161, 170 to 172, 181 to 183 are embodied.
  • the present invention is not limited to the above-described embodiment, and includes various modifications.
  • the above-described embodiment has been described in detail in order to explain the present invention in an easy-to-understand manner, and is not necessarily limited to those having all the described configurations.
  • it is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment and it is also possible to add the configuration of another embodiment to the configuration of one embodiment.
  • the performance DB 101 is provided in the operation management support devices 1, 1a to 1d.
  • the present invention is not limited to this, and the actual result DB 101 may be installed on the cloud or the like, and the operation management support devices 1, 1a to 1d may acquire the actual data 201 from the actual result DB 101 installed on the cloud.
  • each part 110 to 112, 110A to 112A, 110B to 112B, 120 to 123, 120A to 123A, 120B to 123B, 130 to 132, 141, 151, 152, 161 and 170 to 172. 181 to 183, actual results DB101 and the like may be realized by hardware, for example, by designing a part or all of them by an integrated circuit or the like. Further, as shown in FIG. 15, each of the above-described configurations, functions, and the like may be realized by software by interpreting and executing a program in which a processor such as a CPU 402 realizes each function.
  • a memory 401 In addition to storing information such as programs, tables, and files that realize each function in HD (Hard Disk), a memory 401, a recording device such as SSD (Solid State Drive), or an IC (Integrated Circuit) card It can be stored in a recording medium such as an SD (Secure Digital) card or a DVD (Digital Versatile Disc). Further, in each embodiment, the control lines and information lines are shown as necessary for explanation, and not all the control lines and information lines are necessarily shown in the product. In practice, almost all configurations can be considered interconnected.

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Automation & Control Theory (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

La présente invention est caractérisée en ce qu'elle comprend : une unité de construction de modèle de prédiction (111) qui génère un modèle de prédiction (211) pour calculer une valeur de commande de prédiction (y1), à l'aide de données de mesure réelles (203) et de données de commande réelles (202) afin d'effectuer une gestion de fonctionnement efficace de l'équipement ; une unité de calcul de valeur de commande de prédiction (112) qui calcule la valeur de commande de prédiction (y1) par application de données de mesure (204) entrées à partir d'une usine (PL) dans le modèle de prédiction (211) ; une unité de construction de modèle de classification (121) qui génère un modèle de classification (221), par classification des données de mesure réelles (203) dans une catégorie prédéterminée ; une unité de calcul de résultat de classification (122) qui sélectionne une catégorie correspondant aux données de mesure (204) ; une unité de traitement d'évaluation de fiabilité (130) qui calcule la fiabilité (232) de la valeur de commande prédite (y1), par comparaison de la valeur de commande de prédiction (y1) calculée par l'unité de calcul de valeur de commande de prédiction (112) avec la valeur de commande réelle des données de commande réelles (202) dans la catégorie ; et une unité de traitement d'affichage qui affiche la fiabilité calculée (232) sur une unité d'affichage.
PCT/JP2020/042333 2020-01-08 2020-11-12 Dispositif de support de gestion de fonctionnement et procédé de support de gestion de fonctionnement WO2021140742A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2020001608A JP7473343B2 (ja) 2020-01-08 2020-01-08 運転管理支援装置及び運転管理支援方法
JP2020-001608 2020-01-08

Publications (1)

Publication Number Publication Date
WO2021140742A1 true WO2021140742A1 (fr) 2021-07-15

Family

ID=76787861

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2020/042333 WO2021140742A1 (fr) 2020-01-08 2020-11-12 Dispositif de support de gestion de fonctionnement et procédé de support de gestion de fonctionnement

Country Status (2)

Country Link
JP (1) JP7473343B2 (fr)
WO (1) WO2021140742A1 (fr)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2023173706A (ja) * 2022-05-26 2023-12-07 三菱重工業株式会社 状態量予測装置、状態量予測方法、状態量予測システム、及び状態量予測システムの制御方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH03216705A (ja) * 1990-01-23 1991-09-24 Hitachi Ltd プラント運転支援方法及び装置
JP2004310492A (ja) * 2003-04-08 2004-11-04 Nippon Steel Corp プロセスの状態類似事例検索方法及び状態予測方法、並びにコンピュータ読み取り可能な記憶媒体
JP2016045799A (ja) * 2014-08-25 2016-04-04 富士電機株式会社 予測モデル生成装置、予測モデル生成方法及びプログラム
WO2016208315A1 (fr) * 2015-06-22 2016-12-29 株式会社日立製作所 Dispositif de diagnostic d'installation et procédé de diagnostic d'installation
JP2018181052A (ja) * 2017-04-17 2018-11-15 富士通株式会社 モデル同定装置、予測装置、監視システム、モデル同定方法および予測方法

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH04138502A (ja) * 1990-09-29 1992-05-13 Nippon Telegr & Teleph Corp <Ntt> 操作量逐次予測運転システムおよび制御量逐次予測自動制御システム

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH03216705A (ja) * 1990-01-23 1991-09-24 Hitachi Ltd プラント運転支援方法及び装置
JP2004310492A (ja) * 2003-04-08 2004-11-04 Nippon Steel Corp プロセスの状態類似事例検索方法及び状態予測方法、並びにコンピュータ読み取り可能な記憶媒体
JP2016045799A (ja) * 2014-08-25 2016-04-04 富士電機株式会社 予測モデル生成装置、予測モデル生成方法及びプログラム
WO2016208315A1 (fr) * 2015-06-22 2016-12-29 株式会社日立製作所 Dispositif de diagnostic d'installation et procédé de diagnostic d'installation
JP2018181052A (ja) * 2017-04-17 2018-11-15 富士通株式会社 モデル同定装置、予測装置、監視システム、モデル同定方法および予測方法

Also Published As

Publication number Publication date
JP7473343B2 (ja) 2024-04-23
JP2021111057A (ja) 2021-08-02

Similar Documents

Publication Publication Date Title
US7213174B2 (en) Provision of process related information
Idri et al. Fuzzy analogy: A new approach for software cost estimation
Vistro et al. The cricket winner prediction with application of machine learning and data analytics
Bai et al. Inexact genetic algorithm approach to target values setting of engineering requirements in QFD
CN117875724A (zh) 一种基于云计算的采购风险管控方法及系统
CN116485020B (zh) 一种基于大数据的供应链风险识别预警方法、系统及介质
WO2021140742A1 (fr) Dispositif de support de gestion de fonctionnement et procédé de support de gestion de fonctionnement
Hsu et al. The hybrid of association rule algorithms and genetic algorithms for tree induction: an example of predicting the student course performance
Cuadrado-Gallego et al. Analogies and differences between machine learning and expert based software project effort estimation
Koulinas et al. A machine learning-based framework for data mining and optimization of a production system
CN115858606A (zh) 时序数据的异常检测方法、装置、设备及存储介质
Ma et al. Inductive data mining based on genetic programming: Automatic generation of decision trees from data for process historical data analysis
CN117216702A (zh) 变电设备参数处理方法、装置及电子设备
Ashtaiwi Artificial intelligence is transforming the world development indicators
Pattnaik et al. A survey on machine learning techniques used for software quality prediction
Rotovei et al. A methodology for improving complex sales success in CRM Systems
Sallehuddin et al. Forecasting small data set using hybrid cooperative feature selection
Titova et al. Non-destructive diagnosis of food products using neural-genetic algorithm
Jenhani et al. Sim-pdt: A similarity based possibilistic decision tree approach
Leyva-López et al. A multiobjective evolutionary algorithm for deriving a final ranking from a medium-sized fuzzy outranking relation
Kempowsky et al. Supervision of complex processes: strategy for fault detection and diagnosis
Kau et al. An overview and comparison of machine learning techniques
AU2021280953B2 (en) An industrial process model generation system
Ioniță et al. Intelligent system for diagnosis of a three-phase separator
Finlay Types of predictive models

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20911719

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20911719

Country of ref document: EP

Kind code of ref document: A1