Nuclear power plant event multi-dimensional correlation and correlation judgment system and judgment method

Abstract

The invention relates to the field of nuclear event identification, in particular to a nuclear power plant event multi-dimensional association and correlation judgment system and judgment method. The system comprises an information extraction module for extracting event text data; the information processing module preprocesses the event text data extracted by the data extraction module and stores the preprocessed event text data into a structured event library; the structured event library is used for indexing the stored event text data; the data judgment module performs semantic similarity calculation on the specific event and the event of the structured event library to obtain a similarity result of each dimension; and the similarity calculation module performs weighted calculation on the similarity result of the data judgment module to obtain a similarity score of the event in the structured event library and the specific event. According to the method, the similarity between the event in the event library and the specific event in each dimension and after weighting is calculated, and a measurable judgment standard is given, so that the event correlation judgment work is more intelligent, efficient and accurate.

Description

Technical Field

[0001] The present invention relates to the field of nuclear event identification, and in particular to a multi-dimensional association and correlation judgment system and method for nuclear power plant events. Background Art

[0002] Identifying whether there are recurring events and common problems is a very important part of the operation of the nuclear power plant experience feedback system. Nuclear power plants continuously improve the effectiveness of experience feedback by finding recurring events and common problems, deeply analyzing their causes and implementing targeted corrective actions. Generally, each nuclear power plant can generate tens of thousands of various events a year, and how to comprehensively and accurately establish effective connections between events in a large number of events is the basis for identifying recurring events and common problems.

[0003] In the prior art, in order to find events that may be related to a specific event, nuclear power plant personnel need to manually input some search information in the event library and manually screen the search results. However, this process has limitations. First, when setting the search conditions, personnel can only perform matching queries based on the search condition fields preset by the information system. The search dimension is single, and the comprehensiveness and convenience of information retrieval are insufficient; second, the workload of searching and screening events is large. Personnel often need to set search conditions for specific events one by one, and screen them one by one in the large number of retrieved results. This method is inefficient and prone to omissions; third, judging whether an event is related or not is highly dependent on the work experience of personnel. Relevant personnel may misjudge the relevance of events due to information interference, negligence and other factors; finally, the above-mentioned method of judging the relevance of events is relatively subjective and lacks quantifiable standards. It is not easy to dynamically adjust the relevance threshold between events according to the actual management of nuclear power plants, thereby promoting continuous improvement of nuclear power plant management. Summary of the invention

[0004] The present invention provides a nuclear power plant event multi-dimensional association and correlation judgment system and judgment method, which are used to solve the problem that the screening of nuclear power plant recurrence events in the prior art relies on manual work, resulting in incomplete retrieval, relatively subjective and lack of quantifiable standards.

[0005] The technical solution of the present invention is as follows:

[0006] The invention proposes a nuclear power plant event multi-dimensional association and correlation judgment system, which comprises an information extraction module, a structured event library, an information processing module, a data judgment module and a similarity calculation module, wherein the information extraction module extracts event text data; the information processing module pre-processes the event text data extracted by the data extraction module, and stores the pre-processed event text data in the structured event library; the structured event library is used to store the pre-processed event text data, and index the stored event text data; the data judgment module performs semantic similarity calculation on the event text data of a specific event extracted by the information extraction module and pre-processed by the information processing module and the event text data of the structured event library, and obtains similarity results of each dimension between the specific event and other events in the structured event library; the similarity calculation module sets weight coefficients of each dimension, performs weighted calculation on the similarity results of the data judgment module, and obtains similarity scores between the events in the structured event library and the specific event.

[0007] In some embodiments, the information extraction module performs extraction specifically by directly extracting structured fields of event text data and extracting key points using file summary technology for unstructured fields of event text data.

[0008] In some embodiments, the preprocessing of the information processing module specifically includes: removing interference factors in the event text data that may affect the event relevance, and using a professional vocabulary and a synonym vocabulary to perform standardized replacement on the words in the event text data.

[0009] In some embodiments, the data judgment module uses a semantic similarity model pre-trained with a nuclear power corpus to perform multi-dimensional similarity calculations. The data judgment module compares specific events with other events in the structured event library in dimensions including: subject, description, failure point, cause, and corrective action.

[0010] In some embodiments, the data judgment module has the function of setting a determination threshold for event correlation. After setting the determination threshold for event correlation, the data judgment module compares the similarity score with the determination threshold and screens events whose similarity scores are greater than the determination threshold.

[0011] The present invention proposes a method for multi-dimensional association and correlation judgment of nuclear power plant events, the method comprising:

[0012] Step 1: The information extraction module extracts the event text data of the system;

[0013] Step 2: The information processing module preprocesses the event text data extracted by the data extraction module, and stores the preprocessed event text data into the structured event library;

[0014] Step 3: The information extraction module extracts the event text data of the specific event, and the information processing module pre-processes the event text data of the specific event;

[0015] Step 4: The data judgment module calculates the multi-dimensional similarity between the specific event and each event in the structured event library;

[0016] Step 5: The similarity calculation module sets the weights and judgment thresholds for each dimension, and obtains the similarity scores between the specific event and each event in the structured event library through weighted calculation.

[0017] Step 6: The similarity calculation module compares the similarity scores of the specific event and each event in the structured event library with the judgment threshold to screen out events with higher relevance.

[0018] In some embodiments, the event text data in step one comes from a nuclear power plant experience feedback information system.

[0019] In some embodiments, step four specifically includes: after the event text data of the specific event is preprocessed, the data judgment module uses a semantic similarity model pre-trained with a nuclear power corpus to calculate the semantic similarity of the event text data of the specific event with the event text data of other events in the structured event library in the corresponding five dimensions of subject, description, failure point, cause, and corrective action.

[0020] In some embodiments, the screening of events with higher correlation in step six is ​​specifically as follows: if the similarity score of the events in the structured event library is greater than a determination threshold, the events in the structured event library are considered to be high-correlation events.

[0021] The implementation of the present invention has the following beneficial effects

[0022] The present invention proposes a multi-dimensional association and correlation judgment system for nuclear power plant events. A structured event library is constructed through data processing of information extraction modules and information processing modules. After decomposing events into multiple dimensions, the data judgment module and similarity calculation module are used to automatically calculate the similarity of events in the event library and monitored events in each dimension and after weighting, and a measurable judgment standard is given, making the event correlation judgment work more intelligent, efficient and accurate. The problem of incomplete and subjective retrieval caused by manual reliance on recurring event screening in nuclear power plants is avoided. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A schematic diagram of a method for multi-dimensional association and correlation judgment of nuclear power plant events proposed in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The technical solution of the present invention is described clearly and completely below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] like Figure 1 As shown in the figure, the present invention proposes a multi-dimensional association and correlation judgment system for nuclear power plant events. The system includes an information extraction module, a structured event library, an information processing module, a data judgment module and a similarity calculation module.

[0026] The information extraction module uses data extraction tools such as Kettle and related interfaces to extract event text data from the nuclear power plant experience feedback information system according to the rules. The specific rules are: for the information that can be directly extracted from the structured fields in the experience feedback information system, the structured fields include: CR number, subject, description, occurrence date, discovery date, plant name, unit number, CR level, action number, corrective action, etc. For the unstructured fields in the experience feedback information system, the information extracted from the event report document using technologies such as text summarization, the unstructured fields include: failure point, direct cause, root cause, contributing cause, etc.

[0027] The information processing module preprocesses the event text data extracted by the data extraction module and stores the preprocessed event text data in the structured event library. Preprocessing includes: preprocessing the extracted event text data according to the rules, including eliminating interference factors that may affect the relevance of the event (such as the unit / department where the incident occurred, date, etc.), and using professional thesaurus and synonyms to standardize and replace the event text data.

[0028] The structured event library is used to store processed event text data and index the stored event text data to improve the efficiency of subsequent data retrieval.

[0029] The data judgment module uses a semantic similarity model pre-trained with the nuclear power corpus to calculate the semantic similarity between the event text data extracted by the information extraction module and pre-processed by the information processing module and the structured event library, and obtains the similarity results of the specific event and other events in the structured event library in terms of subject, description, failure point, cause, corrective action and other dimensions.

[0030] The similarity calculation module sets the weight coefficient of each dimension in the overall similarity calculation result of the event according to the importance of each dimension of the event, performs weighted calculation on the similarity result of the data judgment module, and obtains the similarity score between the event in the structured event library and the specific event. At the same time, according to the actual management of the nuclear power plant, the similarity calculation module sets an adjustable event correlation judgment threshold as a standard for judging whether the monitored event has a high correlation with other events in the database. The similarity score of each event in the database and the specific event is compared with the judgment threshold. When the similarity score of an event exceeds the threshold, the event is judged to have a high correlation with the monitored event. When multiple events have a high correlation with the monitored event, the computer system issues an early warning, indicating that the monitored event may recur or have common problems, and relevant professionals of the nuclear power plant intervene to analyze and deal with the cause.

[0031] The present invention proposes a method for multi-dimensional association and correlation judgment of nuclear power plant events, characterized in that the method comprises:

[0032] Step 1: The information extraction module extracts the event text data from the nuclear power plant experience feedback information system; among them, the existing structured fields in the information system can be directly extracted, such as CR number, subject, description, occurrence / discovery date, plant name, unit number, CR level, action number, action description, etc.; the unstructured fields in the information system are extracted using text summarization and other technologies to extract key points, such as failure point, direct cause, root cause, contributing cause, etc.

[0033] Step 2: The information processing module preprocesses the event text data extracted by the data extraction module. The preprocessing includes removing interference factors that may affect the relevance of the event text data, including information such as units, departments, and time involved in the text description. Use professional vocabulary and synonyms to standardize and replace event text data. Use nuclear power professional vocabulary, general vocabulary, synonyms, etc. to standardize, standardize or replace non-standard expressions and the same pronouns in the text information. Store the preprocessed event text data in the structured event library and perform text segmentation and vectorization, and index the database to ensure the operating efficiency when performing subsequent multi-dimensional similarity calculations of events.

[0034] Step 3: The information extraction module extracts the event text data of the specific event, and the information processing module preprocesses the event text data of the specific event. The extraction and processing of the event text data of the specific event are the same as the method of the event text in the nuclear power plant experience feedback information system in step 1 and step 2.

[0035] Step 4: The data judgment module calculates the multi-dimensional similarity between the specific event and each event in the structured event library; after preprocessing the specific event, the data judgment module uses the semantic similarity model pre-trained by the nuclear power corpus to calculate the semantic similarity of the specific event with other events in the structured event library in the corresponding dimensions of topic (T), description (D), failure point (F), cause (R), corrective action (Ca), etc., and obtains the similarity results of each dimension between the monitored event and other events in the database, namely, topic similarity (Ti), description similarity (Di), failure point similarity (Fi), cause similarity (Ri), and corrective action similarity (Cai). The semantic similarity model pre-trained by the nuclear power corpus can be trained using the Bert model.

[0036] Step 5: The similarity calculation module sets the weights and judgment thresholds of each dimension, and obtains the similarity scores between the specific event and each event in the structured event library through weighted calculation; according to the importance of each dimension of the event and the actual management, the weight coefficients x1 to x5 of the dimensions such as subject, description, failure point, cause, and corrective action in the overall similarity calculation results of the event are set, where the sum of the weight coefficients is 1, and the similarity results of each dimension are weighted to obtain the similarity score Ei of each event in the structured event library and the specific event. The calculation method is as shown in Formula 1.

[0037] Ti*x1+Di*x2+Fi*x3+Ri*x4+Cai*x5=Ei (1)

[0038] Step 5: The similarity calculation module compares the similarity scores of specific events and each event in the structured event library with the judgment threshold, and selects events with higher correlation. When the similarity score of an event exceeds the judgment threshold, the event is judged to have a high correlation with the monitored event. When there are 3 events (the number can be adjusted dynamically) with a high correlation with the monitored event, the computer system issues an early warning, indicating that the monitored event may recur or have common problems, and relevant professionals of the nuclear power plant intervene to analyze and deal with the cause.

[0039] The above embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present invention. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the attached claims.

Claims

1. A multi-dimensional association and correlation judgment system for nuclear power plant events, characterized in that: The system comprises an information extraction module, a structured event library, an information processing module, a data judgment module and a similarity calculation module, wherein the information extraction module extracts event text data; the information processing module pre-processes the event text data extracted by the data extraction module, and stores the pre-processed event text data in the structured event library; The structured event library is used to store preprocessed event text data and index the stored event text data; the data judgment module calculates the semantic similarity between the event text data of the specific event extracted by the information extraction module and preprocessed by the information processing module and the event text data of the structured event library, and obtains the similarity results of each dimension between the specific event and other events in the structured event library; The similarity calculation module sets a weight coefficient for each dimension, performs weighted calculation on the similarity result of the data judgment module, and obtains a similarity score between the event in the structured event library and the specific event.

2. A nuclear power plant event multi-dimensional association and correlation judgment system according to claim 1, characterized in that: The information extraction module specifically performs extraction by directly extracting structured fields of event text data and extracting key points of unstructured fields of event text data using document summary technology.

3. A nuclear power plant event multi-dimensional association and correlation judgment system according to claim 2, characterized in that: The preprocessing of the information processing module is specifically: removing interference factors that may affect the relevance of events in the event text data, and using professional thesaurus and synonyms to standardize and replace words in the event text data.

4. A nuclear power plant event multi-dimensional association and correlation judgment system according to claim 1, characterized in that: The data judgment module uses a semantic similarity model pre-trained with a nuclear power corpus to perform multi-dimensional similarity calculation. The dimensions of the data judgment module comparing a specific event with other events in the structured event library include: subject, description, failure point, cause and corrective action.

5. A nuclear power plant event multi-dimensional association and correlation judgment system according to claim 4, characterized in that: The data judgment module has the function of setting a determination threshold of event correlation. After setting the determination threshold of event correlation, the data judgment module compares the similarity score with the determination threshold and screens out events whose similarity scores are greater than the determination threshold.

6. A method for determining multi-dimensional association and correlation of nuclear power plant events according to any one of claims 1 to 5, characterized in that: The method comprises: Step 1: The information extraction module extracts the event text data of the system; Step 2: the information processing module preprocesses the event text data extracted by the data extraction module, and stores the preprocessed event text data into a structured event library; Step 3: the information extraction module extracts the event text data of the specific event, and the information processing module pre-processes the event text data of the specific event; Step 4: The data judgment module performs multi-dimensional similarity calculation between the specific event and each event in the structured event library; Step 5: The similarity calculation module sets the weights of each dimension and the judgment threshold, and obtains the similarity scores between the specific event and each event in the structured event library through weighted calculation. Step 6: The similarity calculation module compares the similarity scores of the specific event and each event in the structured event library with the determination threshold, and selects events with higher relevance.

7. A method for multi-dimensional association and correlation judgment of nuclear power plant events according to claim 6, characterized in that: The event text data in step 1 comes from the nuclear power plant experience feedback information system.

8. A method for multi-dimensional association and correlation judgment of nuclear power plant events according to claim 6, characterized in that: The step four specifically includes: after the event text data of the specific event is preprocessed, the data judgment module uses a semantic similarity model pre-trained with a nuclear power corpus to calculate the semantic similarity of the event text data of the specific event with the event text data of other events in the structured event library in five dimensions: corresponding subject, description, failure point, cause, and corrective action.

9. A method for multi-dimensional association and correlation judgment of nuclear power plant events according to claim 6, characterized in that: The specific method of screening events with higher correlation in step 6 is: if the similarity score of the events in the structured event library is greater than the determination threshold, the events in the structured event library are considered to be high-correlation events.

Images