JP7377670B2 - Plant operation support system, plant operation support method - Google Patents

Description

The present disclosure relates to a plant operation support system and a plant operation support method.

Various events occur in a plant that is operated by combining a large number of mechanical devices. Events include minor deviations from normal device behavior, abnormalities, and failures. When an event occurs, it is necessary to respond to that event. When an event occurs in a plant, modulations that occur in response to the event may occur before the event occurs. Patent Document 1 describes a learning device that uses machine learning to extract words corresponding to input words.

JP 2017-204219 Publication

Modulation is a factor in the events that occur. Here, since there are many devices in a plant, when identifying an event from factors using a manual or the like, various inspections are required to identify the event. Moreover, one event may occur as a result of a chain or superposition of multiple factors, and it may be difficult to identify the event. Therefore, in plants where various events occur, a lot of experience is required to identify events from factors. Also, if various points are checked based on factors, it takes time to identify the cause. As described in Patent Document 1, words related to the input keyword can be extracted by using a language analysis machine learning section, but there are cases where it is not possible to appropriately narrow down events based on factors. be.

In order to solve the above problems, the present disclosure aims to provide a plant operation support system and a plant operation support method that can appropriately identify events occurring in a plant.

The present disclosure provides a guidance database that stores information on past events that have occurred in a plant and at least one factor that causes the occurrence of the event, divided into categories based on the relationship with the event; an input section in which sensing information based on the state can be input in free text; and an input section for inferring that something is occurring in the plant based on the sensing information input to the input section and each factor of the guidance database. A plant operation support system is provided, comprising: an output unit that outputs a question based on a factor associated with an article event.

The present disclosure also includes a step of creating a guidance database in which an event that occurs in a plant and at least one factor that causes the occurrence of the event is categorized based on the relationship with the event, and a step of creating a guidance database that is input in free text. a step of matching the sensed information with each factor of the guidance database, and a step of determining the state of the plant based on the matching result. A plant operation support method is provided, which includes the steps of inferring an article event and outputting a question based on a factor associated with the inferred event.

With the above configuration, it is possible to appropriately identify the cause of an event that occurs in a plant.

FIG. 1 is a block diagram showing an example of a plant operation support system. FIG. 2 is a table showing an example of the guideline database. FIG. 3 is a schematic diagram showing an example of propositions and causal relationships between events and factors in the guideline database. FIG. 4 is a table showing an example of the history database. FIG. 5 is a flowchart showing an example of the operation of the plant operation support system. FIG. 6 is a flowchart showing an example of the operation of the plant operation support system. FIG. 7A is an explanatory diagram for explaining the operation of the plant operation support system. FIG. 7B is an explanatory diagram for explaining the operation of the plant operation support system. FIG. 8 is a flowchart showing an example of the operation of the plant operation support system. FIG. 9 is an explanatory diagram for explaining the operation of the plant operation support system. FIG. 10 is an explanatory diagram for explaining the operation of the plant operation support system.

Embodiments according to the present disclosure will be described in detail below based on the drawings. Note that the present invention is not limited to this embodiment. Furthermore, the constituent elements in the embodiments described below include those that can be easily replaced by those skilled in the art, or those that are substantially the same. Furthermore, the constituent elements described below can be combined as appropriate, and if there are multiple embodiments, it is also possible to combine each embodiment.

<Plant operation support system>
FIG. 1 is a block diagram showing an example of a plant operation support system. The plant operation support system 10 according to the present embodiment infers events and factors occurring in the plant based on the input free text information, and outputs answers based on the sensed information of the events and factors as a result of the analogy. do. Here, the plant includes a plant where a nuclear reaction is performed, such as a nuclear power generation facility. Here, the event includes a slight deviation from normal equipment behavior, equipment abnormality, component failure, and the like. A factor is a behavior that occurs in equipment or equipment due to the occurrence of an event. Sensing information is information that a person such as an operator who manages and operates a plant senses from events and factors. Sensing information includes alarms and numerical information that can be sensed with the five senses, and the behavior of equipment and equipment when a human performs an operation to reproduce an event or factor.

The plant operation support system 10 is, for example, permanently installed in a control room of a plant or installed in a portable hardware resource. The plant operation support system 10 may be installed at a location away from the plant, for example, at a facility of a service provider that performs maintenance of various plants. The plant operation support system 10 includes an input section 12, an output section 14, a communication section 16, a control section 18, and a storage section 20.

The input unit 12 includes an input device such as a keyboard and mouse, a touch panel, or a microphone that collects speech from an operator, and outputs a signal corresponding to an operation performed by the operator on the input device to the control unit 18. The input unit 12 has a function that allows the operator to input characters in free text, or to convert voice data acquired through the microphone into text and use it as input information. The output unit 14 includes a display device such as a display, and displays a screen containing various information such as characters and graphics based on a display signal output from the control unit 18. The communication unit 16 is a communication interface that communicates with external devices to send and receive information. The communication unit 16 transmits various data and programs acquired through communication with external devices to the control unit 18 and stores them in the storage unit 20. Note that the communication unit 16 may be connected to an external device via a wired communication line or may be connected to an external device via a wireless communication line. Furthermore, if the plant operation support system 10 is inside a plant and is cut off from the outside, it may be connected only to external equipment inside the plant.

The control unit 18 includes an integrated circuit such as a CPU (Central Processing Unit) and a memory serving as a work area, and executes various processes by executing various programs using these hardware resources. Specifically, the control unit 18 reads a program stored in the storage unit 20, loads it in the memory, and causes the CPU to execute instructions included in the program loaded in the memory, thereby executing various processes. The control unit 18 includes an input analysis unit 32, a matching unit 34, an answer creation unit 36, and a database update unit 38. Before explaining each part of the control section 18, the storage section 20 will be explained.

The storage unit 20 is composed of a nonvolatile storage device such as a magnetic storage device or a semiconductor storage device, and stores various programs and data. The storage unit 20 includes a control program 42, a plant database 44, a guidance database 46, and a response history database 48. The programs stored in the storage unit 20 include a control program 42 that supports plant operation. Note that the storage unit 20 may install the control program 42 by reading the control program 42 recorded on a recording medium, or may install the control program 42 by reading the control program 42 provided on the network. May be installed. Note that the control program 42 may be created using machine learning.

Furthermore, the data stored in the storage unit 20 includes a plant database 44, a guidance database 46, and a response history database 48 used for processing by the control program 42. The plant database 44 includes various information about plants to be supported. Specifically, it includes the equipment configuration of the plant, operating conditions, manual for operating the plant, etc.

FIG. 2 is a table showing an example of the guidance database 46. The guidance database 46 is data used when the matching unit 34 executes matching processing. The guidance database 46 includes guidance data 46a, 46b, 46c, and 46d for each piece of equipment that makes up the plant. Note that although four pieces of guidance data are shown in this embodiment, the number is not limited. Examples of equipment include pumps and their electric motors classified as rotating equipment, valves classified as general structural parts, and fans, compressors, etc. that fall under either of the above two classifications. The equipment guidance data 46a, 46b, 46c, and 46d of the guidance database 46 store events, factors corresponding to the events, and relationships between sensing information of the event and sensing information of various factors. As shown in FIG. 2, the guidance data 46a to 46d are data in which the relationship between an event and the sensing information of a factor is summarized for each event. Sensing information of at least one factor is associated with one event. Further, factors are classified into direct factors, indirect factors, background factors, and expansion factors. The relationship between direct factors, indirect factors, and background factors increases in the order of direct factors, indirect factors, and background factors. The direct factor is the cause of the event occurring in the equipment where the operator sensed the modulation. Indirect factors are factors that contribute to the sensing of direct factors. Further, indirect factors are factors that are necessary conditions for direct factors. If an indirect factor is detected, there is a possibility that a direct factor is present. Background factors are causes of events occurring in equipment other than the equipment in which the operator sensed the modulation.

FIG. 3 is a schematic diagram showing an example of propositions and causal relationships between events and factors in the guidance database. FIG. 3 shows one event among a plurality of events included in the guidance database 46. As shown in FIG. 3, the event 102 is associated with direct factors 104 and 106, indirect factors 108 and 110, and background factors 112 and 114. Further, a true-opposite-opposite relationship is defined between an event and a factor, and between a factor and a factor. A contra-conjunction relationship refers to a one-to-one relationship, that is, a relationship that is a necessary and sufficient condition. In the case of the relationship between an event and a factor, if there is an event, there is also a direct sensing of the factor, and if there is no event, there is no direct sensing of the factor. Note that although the example of FIG. 3 does not show a magnification factor, the relationship between an event and a factor may include a magnification factor. The expansion factor becomes information added to the corresponding factor. Furthermore, an expansion factor is an additional factor that increases the certainty that an event has a relationship with a corresponding direct (including indirect) factor or background factor. Direct factor 104 is in a conjunctive relationship with event 102 . The direct factor 106 has a negative relationship with the event 102. The direct factor 104 and the direct factor 106 are related to the same equipment, and it is possible to determine whether it is the direct factor 104 or the direct factor 106 depending on whether the information for the event 102 is a contraposition or a tail. . Further, the direct factor 104 and the indirect factor 108 have a true dependent relationship. The indirect factor 108 and the indirect factor 110 are in a secret relationship. Direct factor 104 and background factor 112 have an inverse relationship. The direct factor 104 and the background factor 114 are in a dependent relationship. Indirect factor 110 and background factor 114 have a true dependent relationship.

Even if the guidance database 46 is input and created by a person who manages and operates this system based on information on past events (information on events such as minor deviations from normal device behavior, abnormalities, and failures), The database update unit 38 may perform the test and update using machine learning. Alternatively, the test may be performed using machine learning, and the test results may be checked and updated by a person who manages and operates the system.

FIG. 4 is a table showing an example of the response history database 48. The response history database 48 accumulates various information processed by the input analysis section 32, the matching section 34, and the answer creation section 36. The response history database 48 stores inputs detected by the plant operation support system 10 and determination results in association with each other. As shown in FIG. 4, the response history database 48 includes information on factors input in one judgment process, events judged based on the information, and information such as evaluations by operators related to the judged events. It will be done. Specifically, the input text (free text), the words extracted from the input text (phrase information), the associated events, the associated factors, the answer, the input for the answer, the previous answer and the input Contains history. The previous answer is the answer that was output from the output unit before the text to be input this time (question-format guidance that is the cause of the text to be input this time).

Next, each part of the control section 18 will be explained. The input analysis unit 32 analyzes information input to the input unit 12. Specifically, the operator's free text detected by the input unit 12 is cleansed by morphological analysis, etc., and phrase information containing factors necessary for identifying the cause is extracted. Cleansing is the process of omitting words and phrases that are unnecessary for identifying the cause. The phrase information is a word. Further, it is preferable that the word/phrase information includes a word/phrase that expresses the relationship between the words/phrases included in the free text. Free text is a sentence or colloquial information created by an operator using text input software and selecting characters. The information entered by the operator includes sensed information about the plant. The input analysis unit 32 can use morphological analysis or the like as a method for extracting phrase information from free text.

The matching unit 34 uses the sensed information inputted by the operator, such as free text, and the known events, factors, and their sensed information stored in the storage unit 20 to match word information included in the inputted sensed information. , collation is performed with word information stored in the storage unit 20 that is confirmed when a known factor occurs. The matching unit 34 outputs, as a matching result, information regarding the event that is inferred as a candidate and various factors that the operator senses until the event is determined. The matching unit 34 can use a matching function based on machine learning such as Latent Dirichlet Allocation (LDA), which is a metric text analysis method, or a matching function executed based on various algorithms. The matching unit 34 includes a phrase analysis unit 34a, a similarity adjustment unit 34b, and a transfer learning unit 34c. The phrase analysis unit 34a performs matching processing between the phrase information, the guidance database 46, and the sensed information, and identifies known events and factors with the highest degree of similarity. The similarity adjustment unit 34b adjusts a threshold value for similarity determination used in the matching process executed by the phrase analysis unit 34a. Further, as a result of adjusting the threshold value in the similarity adjustment unit 34b, the matching unit 34 transfers the data if the matching process of the matching unit 34 is not completed or if the answers outputted from the answer creation unit 36 have the same contents more than a predetermined number of times. The matching function is switched using the learning unit 34c.

The answer creation unit 36 creates an answer based on the matching result processed by the matching unit 34. Here, the answer is an answer to an input, but may also be a question for acquiring information necessary for driving support. That is, the answer creation unit 36 creates a question that requests the next input as an answer. The answer creation unit 36 can use a sentence creation function created by machine learning, or a sentence creation function executed based on various algorithms.

The database update unit 38 updates the database used when the plant operation support system 10 performs support. Specifically, the input analysis section 32, the matching section 34, and the response creation section 34 use a response history database 48 that stores the history of processing executed by the input analysis section 32, the matching section 34, and the response creation section 36. The answer creation unit 36 and the guidance database 46 used for the processing executed by the answer creation unit 36 are updated. When updating the database in the database update unit 38, it is determined whether the probability of information related to support in the database used when executing support in the plant operation support system 10 is maintained even after the update, using machine learning in advance before the update. You may also use a method of accepting or rejecting. The database update unit 38 uses, for example, latent Dirichlet allocation (LDA) to generate a topic distribution, performs matching, and identifies candidate events to be associated.

<Plant operation support method>
Next, a plant operation support method executed by the plant operation support system will be described using FIGS. 5 to 6. 5 and 6 are flowcharts showing an example of the operation of the plant operation support system. FIG. 5 is a flowchart showing the entire process of the plant operation support method. FIG. 6 is a flowchart showing a specific example of the processing executed by the matching unit 34 and the answer creation unit 36.

The entire process of the plant operation support method will be explained using FIG. 5. The control unit 18 detects input from the operator (step S12). Input by the operator is performed by free text input. That is, in this embodiment, the operator inputs character information using text input software, rather than inputting by selecting displayed options. The input information is the operating state of the plant and sensed information regarding the plant, as described above.

The control unit 18 decomposes the input information into phrase information and performs matching with the database (step S14). The control unit 18 uses the input analysis unit 32 to analyze the detected free text and extract at least one phrase information. For example, if a 10 degree rise in waste water temperature is input in free text, temperature can be extracted as phrase information, the matching unit 34 can identify the document as having temperature as a topic, and associate 10 degrees with the increase. preferable.

Next, the processing by the matching unit 34, specifically, the processing in step S14, will be explained in more detail using FIG. 6. The processing of the control unit 18 will be described below. The control unit 18 acquires phrase information (step S32). The control unit 18 identifies the directionality of the topic based on the number of times each word appears in all the information input using the matching unit S34 and the bias (bias) of the frequency of appearance of the specific word in all the word information. Then, the topic directionality of each event registered in the guidance database 46 and the factor corresponding to the event is compared, and the degree of similarity is calculated. A person who manages and operates this system determines whether the degree of similarity is greater than or equal to a preset similarity threshold, and if it is greater than or equal to the threshold, it is detected that there is a matching factor.

If the control unit 18 determines that there is a matching factor (Yes in step S34), it specifies the event corresponding to the matching factor as a candidate event (step S36).

The control unit 18 determines whether the specified event is confirmed (step S38). Specifically, in step S38, when a plurality of candidate events are identified, the matching unit 34 determines the event with the highest degree of similarity as the corresponding event.

If the event specified by the input of the operator's evaluation result is confirmed (Yes in step S38), the control unit 18 sets the confirmed event as a matching result (step S40), and ends the process of FIG. 6. At the same time, the operator's input information from steps S32 to S38, which are registered in the response history database, the matching results, and the evaluation results regarding whether they are correct or incorrect are sent to the person who manages and operates this system.

If the identified event cannot be determined based on the operator's input of the evaluation result (No in step S38), the control unit 18 determines the event with the next highest degree of similarity to the identified event as the matching result. Create (step S42). As a result, if the event has been determined (Yes in step S43), the determined event is taken as the matching result (step S40), and the process of FIG. 6 is ended. At the same time, the operator's input information from steps S32 to S38, which are registered in the response history database, the matching results, and the evaluation results regarding whether they are correct or incorrect are sent to the person who manages and operates this system. If No in step S43, the process returns to step S34.

If the control unit 18 determines that there is no factor that is equal to or higher than the preset similarity threshold and that there is no matching factor (No in step S34), the person who manages and operates this system can use the preset matching factor. It is determined whether or not the similarity threshold adjustment for detecting factors has reached a predetermined number of times (step S44). In other words, it is determined whether the similarity threshold can be lowered. When the control unit 18 determines that the similarity threshold can be adjusted (No in step S44), the control unit 18 adjusts the similarity threshold (step S46), that is, sets the threshold to a lower value, and performs the processing in step S34. Proceed to. As a result, the similarity threshold serving as the standard for event identification is lowered, and based on the lower similarity threshold, it is again detected whether there is a matching factor.

If the control unit 18 determines that the similarity threshold cannot be adjusted (Yes in step S44), it switches from the similarity adjustment unit 34b to the transfer learning unit 34c. (Step S48). The transfer learning unit 34c simply detects a corresponding factor based on the similarity between the topic directionality of word information extracted from the sensed information input by the operator and the topic directionality related to known events registered in the guidance database 46. Rather, it sequentially arranges a series of phrase information input by the operator. Similarly, the factors corresponding to each event included in the guidance data 46a to 46d are arranged in order of relevance to the event, and the degree of diachronic consistency between the two arrangements is compared. In other words, events are identified that have a high degree of agreement in the structure of causal relationships (true, opposite, opposite, and even) between events and factors. For example, when an operator inputs a vibration event as sensing information in a pump, there is no registration of the known information in the pump, and the known information in the compressor, which is a different device, and the structure of the causal relationship between the event and the factors are not registered. If the degree of matching is high, the device name is automatically tuned from the compressor to the pump, and a matching result is created (step S50).

If there is a matching factor and the event can be determined (Yes in step S51), the control unit 18 determines the event corresponding to the matching factor as a corresponding event (step S40) and ends the process of FIG. 6. . At the same time, the operator's input information from step S32 to step S51 registered in the response history database, the matching results, and the evaluation results regarding correctness and incorrectness are processed to be sent to the person who manages and operates this system.

If there is no matching factor and the event cannot be determined (No in step S51), the control unit 18 determines that the process in FIG. Step S54), this process ends. The control unit 18 further sends the operator's input information and matching results from step S32 to step S51, which are registered in the response history database, and the evaluation results regarding correctness and incorrectness to the person who manages and operates this system. do.

Returning to FIG. 5, the overall explanation of the plant operation support method will be continued. When the matching result is determined, the control unit 18 creates an answer based on the matching result (step S16). Here, the answer is the answer to the operator's input. For example, if the event has been determined, the control unit 18 responds with event information and items necessary for identifying the event. Further, if the event has not been determined, the control unit 18 outputs a question regarding the factor selected as requiring confirmation based on the guidance database 46 and the identified event as an answer.

The control unit 18 detects an input for an answer (step S18). Inputs for answers include free text, selection of options, input of processing completion information, and the like. When the control unit 18 detects an input to the answer, it determines whether the process is finished (step S20).The control unit 18 determines whether the conditions for identifying the event or terminating the driving support are satisfied based on the input for the answer. If it is determined, it is determined that the process has ended. If the control unit 18 determines that the process has not ended (No in step S20), the process returns to step S14, and the free text detected in step S18 and No in step S20 are used as new inputs to perform matching, answer creation, and response. Detect input to . If the control unit 18 determines that the process has ended (Yes in step S20), it ends this process.

Next, an example of the process executed in the process of FIGS. 5 and 6 will be described using FIGS. 7A and 7B. FIGS. 7A and 7B are examples of screens on which information for the operator is displayed. FIGS. 7A and 7B are examples of the transition of the display screen when an event related to the pump occurs. The screens shown in FIGS. 7A and 7B may be displayed as a window on a part of the display device, or may be displayed on the entire screen. The plant operation system 10 may output the information shown in FIG. 7 in the form of audio, and may input the information as audio.

For example, the plant operation system 10 displays a screen 150 shown in FIG. 7A. On the screen 150, parameters of each part included in the plant are displayed as items. Parameters include the equipment itself, data detected by the equipment, temperature, pressure, flow rate, etc. When an operator detects an abnormality, the operator selects the item for which the abnormality is detected. In FIG. 7A, item 152 in screen 150 is selected. Item 152 is the pump discharge pressure. In this embodiment, the input for item 152 is selective, but pump discharge pressure may also be input by free text input.

The plant operation system 10 displays a screen 154 when an operation on the item 152 is detected. Screen 154 includes a question column 156 and an answer column 158. Screen 154 is a screen for checking the operating status of the pump. In the question field 156, "What is the driving condition?" is displayed. In the answer column 158, answer candidates such as "normal operation", "regular switching", "trial run", and "YYY water injection test" are displayed in a selection format. In this embodiment, periodic switching is selected.

When the plant operation system 10 detects selection during periodic switching, it displays the screen 160. Screen 160 includes an answer column 162. Screen 160 is a screen where the operator inputs sensed information in free text. The answer column 162 is a column into which words and phrases are input in free text. In this embodiment, an input such as "After the XXX pump was started, a negative pressure indication began to appear on the discharge pressure gauge" is detected. The plant operation system 10 decomposes the input text into phrase information. Specifically, word information such as pump, start, after, protruding pressure gauge, and negative pressure instruction is extracted.

The plant operation system 10 displays the screen 164a when an input to the answer column 162 is detected. Screen 164a includes a question column 166a and an answer column 168a. The screen 164a is a screen for confirming information in an interactive manner. In the question column 166a, "Negative pressure even after XXX pump is stopped?" is displayed. The question field 166a is a question based on sensing an event. In the answer field 168a, answer candidates are displayed in a selection format: Probably partially yes, Probably not, Yes, I don't know, and No. In this embodiment, "probably partially yes" is selected. Note that if an answer including "maybe" is input, the user is asked to input information to supplement the question in free text or voice.

The plant operation system 10 displays the screen 164b when the selection of "Probably partially yes" in the answer column 164a is detected. Screen 164b includes a question column 166b and an answer column 168b. The screen 164b is a screen for confirming information in an interactive manner. In the question column 166b, "Will air suction into the pump itself be confirmed after XXX pump is stopped?" is displayed. The question column 166b is a question based on the direct factor itself or sensed information of the direct factor. In the answer column 168b, answer candidates are displayed in a selection format such as "Maybe partially yes," "Maybe not," "Yes," "I don't know," and "No." In this embodiment, "No" is selected.

The plant operation system 10 displays the screen 164c when the selection of "No" in the answer column 164b is detected. The screen 164c includes a question column 166c and an answer column 168c. The screen 164c is a screen for confirming information in an interactive manner. In the question column 166c, "Does the MMM tank water level gauge maintain the maximum water level?" is displayed. Question field 166c is a question based on direct factor sensing. In the answer column 168c, answer candidates are displayed in a selection format such as "Maybe partially yes," "Maybe not," "Yes," "I don't know," and "No." In this embodiment, "probably different" is selected.

When the selection of "Probably not" in the answer column 168c is detected, the plant operation system 10 requests the operator to input a specific amount of change in the water level, for example, the amount of decrease in the scale of the water level gauge, by free text or voice. do. After the input is completed, a screen 164d is displayed. The screen 164d includes a question column 166d and an answer column 168d. The screen 164d is a screen for confirming information in an interactive manner. “Winter?” is displayed in the question field 166d. The question field 166d is a question based on the background factor itself or sensed information of the background factor. In the answer column 168d, answer candidates are displayed in a selection format such as "Maybe partially yes," "Maybe not," "Yes," "I don't know," and "No." In this embodiment, "Yes" is selected.

The plant operation system 10 displays the screen 164e when the selection of "Yes" in the answer column 164d is detected. Screen 164e includes a question column 166e and an answer column 168e. The screen 164e is a screen for confirming information in an interactive manner. In the question field 166e, "Is XX work performed between MMM tanks?" is displayed. The question column 166e is a question based on sensed information of background factors. The answer field 168e displays answer candidates such as "Maybe partially yes," "Maybe not," "Yes," "I don't know," and "No." In this embodiment, "Yes" is selected.

The plant operation system 10 displays a screen 164f when a selection of "Yes" in the answer column 164e is detected. The screen 164f includes a question column 166f and an answer column 168f. The screen 164f is a screen for confirming information in an interactive format. In the question column 166f, "Is the distance between the area around the XXX pump and the MMM tank outside the XXX pump system large?" is displayed. The question field 166f is a question based on the magnification factor itself or sensed information of the magnification factor. In the answer column 168f, answer candidates are displayed in a selection format such as "Maybe partially yes," "Maybe not," "Yes," "I don't know," and "No." In this embodiment, "Yes" is selected.

The plant operation system 10 displays the screen 170 when the selection of "Yes" in the answer column 164f is detected. Screen 170 includes a question column 172 and an answer column 174. The screen 170 displays the identified cause in the question column 172 after the cause has been identified, and displays the countermeasure corresponding to the cause in the answer column 174. Screen 170 is a screen for identifying and confirming events and factors. Question column 172 displays "There is air suction in the MMM tank upstream outside the XXX pump system, and there is a possibility that air has entered the XXX pump." The question field 172 shows an analogy of the cause of the abnormality, and is created based on direct factors, indirect factors, background factors, and sensed information of each factor. A question column 172 displays the analysis results. In the answer column 174, "Please increase the flow rate of lubricating water from the MMM tank to the XXX pump, discharge the air pocket, and then check the instructions on the protruding pressure gauge again." The answer column 174 displays countermeasures.

Next, the updating process of the guidance database 46 executed by the database updating unit 38 will be explained using FIG. 8. FIG. 8 is a flowchart showing an example of the operation of the plant operation support system. The process of updating the guidance database 46 by the database update unit 38 may be executed at predetermined intervals or every time data is accumulated in the response history database 48.

The database update unit 38 reads the response history database 48 (step S72). Next, the database update unit 38 reads the guidance database 46 (step S74). The database update unit 38 performs database verification (step S76). Specifically, the data in the response history database 48 and the data in the guidance database 46 are compared to determine whether there is an event in the data stored in the response history database 48 that does not completely match the guidance database 46. Also, it is determined whether there is a relationship between the event and each factor registered in the guidance database 46.

Based on the matching results, the database update unit 38 selects the topic distribution of the specific event registered in the guidance database 46 that is determined to be most relevant from the topic information accumulated in the response history database 48, It is selected as an event to be tested (step S78).

The database update unit 38 performs the test by comparing the topic distribution of the test target and the topic distribution of the test target accumulated in the response history database, and executes adoption (Yes) or rejection (No) (step S80). For example, the Metropolis Hasting method can be used for the assay. When the database update unit 38 determines to be rejected (No in step S80), the database update unit 38 performs a test (semi-automatically) by comparing the topic analysis group other than the test target and the topic distribution group newly accumulated in the response history database 48, and determines the selection. (Yes) or rejection (No) (step S81). The database update unit 38 can semi-automatically perform the process of step S81 by using the process described below. When the database update unit 38 determines that the application is adopted in step S81 (Yes in step S81), the process proceeds to step S82. When the database update unit 38 determines that the request is rejected in step S81 (No in step S81), the process proceeds to step S84.

When the database update unit 38 determines that the factor is adopted (Yes in step S80), the database update unit 38 specifies the type of factor. That is, it is determined whether the adopted factor is a direct factor, indirect factor, expansion factor, or background factor, and the adopted factor is newly registered and updated in the guidance database 46 (step S82), and the response history database is checked. It is determined whether the process has been completed (step S84). If the database update unit 38 determines that the confirmation has not been completed (No in step S84), the process returns to step S78, executes the process for the next test target, and determines that the confirmation has been completed (Yes in step S84). If so, the process ends.

FIG. 9 is an explanatory diagram for explaining how the guidance database 46 is updated by the database updater 38. FIG. 10 is an explanatory diagram showing that the topic distribution group of the guidance database 46 is formed in a multi-axis space. FIG. 10 shows the distribution of each word for each topic group, with the arbitrary word a, the arbitrary word b, and the degree of deflection σ as the axes. FIG. 9 shows an example of processing when a new factor accumulated in the response history database is associated with event A using the Metropolis Hasting method. As shown in FIG. 9, the database updating unit 38 has a guidance database 46 that is preset based on information on past events, and is provided with an A event area 204 and a B event area 206. The area 204 includes an area 210 formed by phrase information of the factor associated with the A event. In addition, in order for the region 210 to be registered as an A event, based on the number of times each word appears in the distribution 230 of arbitrary words in all information and the tendency of bias of specific words that appear in the information, be identified. Next, the database update unit 38 updates the topic directionality (vector 202), in the region 212, based on (vector 203), it is tested whether regions 212 and 214 on the probability density function space, which is the region to be tested, are adopted or rejected as the A event. For example, the number of occurrences of an arbitrary word included in the region 210 and the number of occurrences of the word in the regions 212 and 214 on the test side that are subject to additional updating are determined by the difference in probability density between the region 210 and the like. Become. In addition, the topic directionality estimated from the relevance of each word included in the area 210 and the topic directionality estimated from the areas 212 and 214 on the test side (area 212 is the vector 203, area 214 is the vector 202) ) is the degree of discreteness from the area distribution representing the A event. As shown in FIG. 10, the topic distribution group maps all word information on the guidance or response history database onto a multi-axis space, and is used to compare topic directivity and calculate similarity. The database update unit 38 determines whether or not the A event can be added to the guidance database by comparing the difference in probability density and the degree of discreteness between the regions 212 and 214 on the test side and the above-mentioned A event. 212 can be tested as rejection. In the example shown in FIG. 9, the region 214 is identified as an event similar to the A event or an event that includes a highly related factor. As a result, the spatial region 204 of the A event is expanded to a region 208 . Thereby, for example, when information regarding the A event or the above-mentioned similar event is newly input by the operator, it is possible to improve the matching rate of similarity by factor comparison. Furthermore, the phrase information in the area 212 can be automatically rejected if it is not in the area group of the A event, and the updating of the guidance database can be saved. In addition, regarding region 212, comparison with multiple region groups of other events other than event A allows for semi-automation of the test of acceptance and rejection, allowing the operator's sensed information to be efficiently updated and reused in the guidance database. You can also do it.

In order to maintain the validity of the above-mentioned verification (acceptance, rejection) based on the Metropolis Hastings Law, etc., the person who manages and operates this system must check the above at regular intervals based on the operator's evaluation information for the judged event, and the above-mentioned To confirm the probability as a verification function of Metropolis Hasting method etc., (1) Benchmarks, (2) A Posteriori Error Estimation (J.T.Oden ) etc. may be used.

After identifying the type of factor, the database update unit 38 registers the corresponding event and factor in the guidance database 46 in association with the type of factor (step S82).

When the database update unit 38 registers in the guidance database 46, it determines whether confirmation of the response history database 48 is completed (step S84). That is, it is determined whether there is any history data that has not been confirmed in the read response history database 48. If the database update unit 38 determines that the confirmation has not been completed (No in step S84), that is, if there is history data that has not been confirmed, the process returns to step S78. If the database update unit 38 determines that the confirmation has been completed (Yes in step S84), that is, if there is no history data that has not been confirmed, the database update unit 38 ends this process.

The plant operation support system 10 uses the guidance database 46 to associate events, factors, event sensing information, and factor sensing information, and to classify factors by type of factor, so that the information input by the operator is The accuracy of responses to free text can be increased. Further, by outputting a question with an answer corresponding to each type of factor, it is possible for the operator to easily detect input regarding the factor. Furthermore, by continuously collecting information in an interactive manner with the operator, the number of occurrences of each input word can be increased, and the similarity between the event sensing information, factor sensing information, and the event can be evaluated with higher accuracy. I can do it. Thereby, it is possible to more appropriately identify an event occurring in the plant, and it is possible to identify the event with high accuracy.

Further, as in this embodiment, by including questions necessary for identifying the event as answers and collecting sensed information of the operator, the event can be identified with a higher probability.

In addition, as in this embodiment, if an event in equipment that corresponds to the operator's input of sensed information cannot be identified, a transfer learning function or the like can be used to temporarily identify an event in another equipment without identifying the event. By setting it arbitrarily, using the matching unit 34 to infer the factor with the highest degree of similarity in other equipment, and once providing an answer as guidance, and obtaining new input from the operator, it is possible to identify any event that has not yet been performed. It is also possible to identify the event as a postponed event, switch to guidance on the factors of the event, and enable identification of the event. By switching the matching unit from the machine learning described above to AI, it is possible to add robustness to event identification.

Furthermore, by updating the guidance database as in the present embodiment, sensing information for new events can be stored in association with factors, making it possible to identify more events. It is also possible to start operation without any data in the guidance database 46. In addition, the guidance database 46 can store the knowledge of the design FMEA of general industrial equipment, even if it is not information on past events (information on events such as minor deviations from normal equipment behavior, abnormalities, and failures). Events and factors, information that can be sensed at the time of occurrence, equipment troubleshooting (how to deal with abnormality detection), etc. can be created and registered by associating them with each other for each mode. Specifically, the design FMEA includes knowledge of failure modes, stress, and strength assumed at the time of design. Here, stress refers to external factors such as usage conditions, environment, frequency of use, load, etc. that any equipment is subjected to, and strength refers to the quality characteristics and strength that any equipment has, taking into account the expected stress. , resistance, and other design specifications. Strength has a safety margin against stress at the time of design, but if the design safety margin is exceeded due to external factors such as deterioration of equipment, frequency of use, and increase in load, there is a possibility of failure. It arises. Therefore, by organizing the above-mentioned stress and strength and the above-mentioned coping methods based on the propositions and causal relationships shown in Fig. 3, even when information on past events is limited and matching processing is difficult, the propositions shown in Fig. 3 , based on the causal relationship, it can be used as information on each factor (direct factor, indirect factor, background factor, expansion factor) registered in the guidance database 46, and the amount of data in the guidance database 46 can also be interpolated.

Furthermore, it is preferable that the matching unit 34 and the response creation unit 36 identify the event and create the response using information in the guidance database 46 that has been updated by testing the response history database 48. This allows more efficient and appropriate processing. As shown in FIG. 8, it is preferable to provide a response history database 48 in order to update the guidance database 46, but it does not necessarily have to be provided.

In the above embodiment, free text and selection format input is used, but the input for the answer created by the answer creation section 36 may be only free text. Further, as described above, in order to further improve the accuracy of plant operation support, it is preferable to enable information on support results to be input and detected in response to answers. In other words, it is preferable to input whether the judgment of the plant operation support system 10 was correct in addition to or instead of the countermeasure method for the specified event, in response to the answer created when the event is determined.

Although the above embodiment has been described assuming that the plant is a nuclear power plant, the present invention is not limited to this, and can also be applied to, for example, a thermal power plant, a chemical plant, etc.

10 Plant operation support system 12 Input unit 14 Output unit 16 Communication unit 18 Control unit 20 Storage unit 32 Input analysis unit 34 Matching unit 34a Phrase analysis unit 34b Similarity adjustment unit 34c Transfer learning unit 36 Answer creation unit 38 Database update unit 42 Control Program 44 Plant database 46 Guidance database 48 Response history database

Claims (14)

a guidance database that stores information on past events that have occurred in the plant and at least one factor that causes the occurrence of the event, divided into categories based on the relationship with the event;
an input unit capable of inputting sensed information based on the state of the plant in free text;
an output unit that outputs a question based on a factor associated with the event that is inferred to be occurring in the plant, based on the sensing information input to the input unit and each factor of the guidance database; and,
Equipped with
The categories include direct factors, indirect factors, expansion factors, and background factors,
The relationship with the event increases in the order of the direct factor, the indirect factor, and the background factor,
The direct factor is the cause of the event occurring in the equipment in which the operator sensed the modulation;
The indirect factor is a factor that contributes to the sensing of the direct factor and is a necessary condition for the direct factor,
The background factor is a cause of an event occurring in a device other than the device in which the operator sensed the modulation,
The expansion factor is information added to a corresponding factor in a plant operation support system. a matching unit that infers the event occurring in the plant based on the sensing information input to the input unit and each factor of the guidance database;
The plant operation support system according to claim 1, further comprising: an answer creation unit that creates a question based on the event matched by the matching unit and a factor associated with the event. When multiple events are inferred as a result of matching by the matching unit, the answer creation unit creates a question in which information regarding other factors of the event that is inferred to have the highest degree of similarity is input. The plant operation support system according to claim 2, wherein questions are answers. 4. The plant operation support system according to claim 2, wherein the answer creation unit outputs a multiple-choice question or a question into which free text can be input as an answer. If the answer cannot be determined based on the output answer, the answer creation unit selects a factor that has the next highest degree of similarity after the factor corresponding to the output answer among other events inferred by the matching unit other than the output answer. The plant operation support system according to any one of claims 2 to 4, wherein a question is created in which information regarding a factor of a high event is input, and the created question is an answer. The matching unit adjusts a matching similarity threshold when determining that a factor cannot be identified between the sensing information input to the input unit and each factor of the guidance database. 4. The plant operation support system according to any one of 4. If the matching unit cannot identify a factor based on the sensing information input to the input unit and each factor in the guidance database, the matching unit uses transfer learning to associate the factor with the event. The plant operation support system according to any one of Item 6. a response history database that stores information input to the input section and information output from the output section;
The plant operation support according to any one of claims 1 to 7, further comprising: a database update unit that stores information input to the input unit and information output by the output unit in the response history database. system. 9. The database updating unit updates the guidance database by associating the event, a factor associated with the event, and a category of the factor based on the input of the response history database. The plant operation support system described in . 10. The plant operation system according to claim 1, wherein the guidance database includes information that associates knowledge of design FMEA of general industrial equipment with methods of handling equipment abnormalities based on propositional causal relationships. support system. The guidance database according to any one of claims 1 to 10, wherein events and factors are defined as opposites or opposites, and direct factors, indirect factors, and background factors are defined as true, opposite, or opposites. plant operation support system. creating a guidance database that categorizes events that occur in the plant and at least one factor that causes the occurrence of the event into categories based on the relationship with the event;
obtaining sensing information input in free text based on the state of the plant;
matching the sensed information with each factor of the guidance database;
The method includes the step of inferring the event occurring in the plant based on the matching result and outputting a question based on a factor associated with the inferred event,
The categories include direct factors, indirect factors, expansion factors, and background factors,
The relationship with the event increases in the order of the direct factor, the indirect factor, and the background factor,
The direct factor is the cause of the event occurring in the equipment in which the operator sensed the modulation;
The indirect factor is a factor that contributes to the sensing of the direct factor and is a necessary condition for the direct factor,
The background factor is a cause of an event occurring in a device other than the device in which the operator sensed the modulation,
In the plant operation support method, the expansion factor is information added to a corresponding factor. In the matching step, if sensing information and factors cannot be matched based on the guidance database, matching is performed by transfer learning,
13. The plant operation support method according to claim 12, wherein the guidance database is updated based on a relationship between an event and a factor determined based on the question and the answer to the question in the transfer learning. According to claim 12 or claim 13, the guidance database is characterized in that a plurality of factors are associated with one event, and the relationship between an event and a factor, or a relationship between a factor and a factor is defined as true, inverse, opposite, or contrapositive. The plant operation support method described.

Images