Intelligent trend analysis method for high-level events of nuclear power plant

Abstract

The invention belongs to the technical field of nuclear power, and particularly relates to an intelligent trend analysis method for high-level events of a nuclear power plant, which comprises the following steps of: 1, carding and artificially coding reason analysis reports of the high-level events to form a trend coding knowledge base of the high-level events; 2, based on a high-level event trend coding knowledge base and a large language model, constructing an automatic coding model of a nuclear power plant high-level event reason analysis report; 3, automatically coding and manually checking the high-level event reason analysis report; 4, automatically identifying an abnormal trend and a common problem existing in a high-level event through a coding result, and outputting the result; and step 5, tracking the identified abnormal trend and common problems. According to the method, the abnormal trend can be automatically identified to identify the problems of equipment, management and the like in the group factory range, so that major events are avoided.

Description

Technical Field

[0001] The present invention belongs to the field of nuclear power technology, and in particular relates to an intelligent trend analysis method for high-level events in a nuclear power plant. Background Art

[0002] At present, there are methods for automatic trend analysis of low-level events in nuclear power plants based on machine learning and deep learning. The technical solution is designed for low-level status reports of nuclear power plants. Its coding system and abnormal trend identification scheme are not applicable to high-level events (A and B levels). In addition, the technical solution uses machine learning and deep learning algorithms, which can obtain relatively accurate results when processing simple text data such as low-level status reports, but has certain limitations when dealing with large-scale, high-dimensional, and irregular text data such as high-level event cause analysis reports. In addition, the coding design and abnormal identification criteria of the automatic trend analysis method for low-level events are not fully applicable to high-level events, and cannot meet the needs of monitoring abnormal trends of high-level events.

[0003] Therefore, it is necessary to develop an intelligent trend analysis method for high-level events to achieve automatic coding of analysis reports of high-level events and identification of abnormal trends, so as to provide support for the experience feedback work of nuclear power plant groups. Summary of the invention

[0004] The purpose of the present invention is to provide an intelligent trend analysis method for high-level events in nuclear power plants. The method is suitable for automatically encoding cause analysis reports of high-level events, and automatically identifying abnormal trends and determining common problems based on trend analysis of high-level event coding, thereby achieving continuous tracking of existing trends.

[0005] The technical solution to achieve the purpose of the present invention is:

[0006] A method for intelligent trend analysis of high-level events in a nuclear power plant, the method comprising:

[0007] Step 1: Sort out and manually code the cause analysis reports of high-level events to form a high-level event trend coding knowledge base;

[0008] Step 2: Based on the high-level event trend coding knowledge base and large language model, an automatic coding model for the cause analysis report of high-level events in nuclear power plants is constructed;

[0009] Step 3: Automatically encode and manually check the cause analysis report of high-level events;

[0010] Step 4: Automatically identify abnormal trends and common problems in high-level events through coding results, and output the results;

[0011] Step 5: Track identified abnormal trends and common problems.

[0012] The step one comprises:

[0013] Step 1.1: Construct a high-level event classification and coding system;

[0014] Step 1.2: Select the high-level event cause analysis report and manually code it according to the high-level event classification and coding system;

[0015] Step 1.3: Enter the high-level event cause analysis report information and coding results into the coding database;

[0016] Step 1.4: Sort out the text data and coding result information of the cause analysis report of high-level events in nuclear power plants, process the text information, and vectorize it using a large language model to build a high-level event trend coding knowledge base.

[0017] In step 1.1, the high-level event codes are divided into equipment, human factors, technology and keyword codes.

[0018] The step 2 includes: using a large language model to perform automatic encoding fine-tuning based on a trend encoding knowledge base to build an automatic encoding model.

[0019] The step three comprises:

[0020] Step 3.1: Extract new high-level event cause analysis report information;

[0021] Step 3.2: Use the large language model to preprocess, segment, and vectorize the report information;

[0022] Step 3.3: Use the automatic coding model to automatically code the high-level event cause analysis report, obtain the coding results, and generate the corresponding coding cause description;

[0023] Step 3.4: Manually check the coding and enter the final coding result into the coding database.

[0024] The cause analysis report information extracted in step 3.1 includes the unstructured information of the description, consequences, cause analysis process and conclusion in the cause analysis report, and the structured information of the subject, unit, system equipment and time in the corresponding status report.

[0025] The fourth step comprises:

[0026] Step 4.1: Set the criteria for determining abnormal trends and common problems;

[0027] Step 4.2: Automatically identify abnormal trends and common problems of high-level events in nuclear power plants based on the judgment criteria and issue warnings, and output abnormal trend data, charts, trend overviews and other information;

[0028] Step 4.3: Manually check and confirm abnormal trends and common problems.

[0029] In step 4.3, if the abnormal trend is confirmed to be an unfavorable trend in the field of power plant experience feedback, management action should be taken to intervene. In this case, a corresponding abnormal trend evaluation conclusion needs to be generated and it is recommended that the management unit issue management actions.

[0030] The step five comprises:

[0031] Step 5.1: Manually determine and include the abnormal trends that deserve attention into tracking;

[0032] Step 5.2: Extract the high-level event trend coding data of the new period and compare it with the existing trends in the abnormal trend tracking database;

[0033] Step 5.3: Output the trend tracking data for the current month;

[0034] Step 5.4: Manually determine whether the abnormal trend has been eliminated. If it has been eliminated, cancel the tracking; if it has not been eliminated, continue to include it in the tracking.

[0035] The beneficial technical effects of the present invention are:

[0036] The present invention provides an intelligent trend analysis method for high-level events in nuclear power plants, which fills the gap in the lack of automatic trend analysis methods and tools for high-level events in the nuclear power field. It uses large language model technology as a basis to realize automatic encoding of cause analysis reports of high-level events in the field of experience feedback from nuclear power plant groups, and can automatically identify abnormal trends to identify equipment, management and other problems within the group, thereby avoiding the occurrence of major events. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a flow chart of a method for intelligent trend analysis of high-level events in a nuclear power plant provided by the present invention. DETAILED DESCRIPTION

[0038] The present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0039] The present invention provides a method for intelligent trend analysis of high-level events in nuclear power plants, comprising a knowledge base formed by combing and encoding the cause analysis reports of high-level events in nuclear power plants, an automatic encoding model of the cause analysis reports of high-level events formed by fine-tuning a large language model, and application data formed by fine-tuning the training of the large model. Specifically, Figure 1 As shown, the following steps are included:

[0040] Step 1: Sort out and manually encode the cause analysis reports of high-level events to form a high-level event trend coding knowledge base.

[0041] Step 1.1: Construct a high-level event classification and coding system.

[0042] At present, the "Group Factory Important Event Coding System" has been designed based on the characteristics of high-level events. The high-level event codes are divided into equipment, human factors, technology and keyword codes. With a problem-oriented approach, it can accurately and completely identify the main characteristics of an important event and the key issues that experience feedback personnel need to pay attention to. An event may involve multiple codes. This design is conducive to ensuring the pertinence and effectiveness of trend analysis and common problem identification.

[0043] Step 1.2: Select a certain number of high-level event cause analysis reports and perform manual coding.

[0044] According to the high-level event classification and coding system, the existing high-level event cause analysis reports were manually coded as the basis for subsequent work. The manual coding work used the function of large language model summarization to reduce the workload of manual coding.

[0045] Step 1.3: Enter the high-level event cause analysis report information and coding results into the coding database.

[0046] The coding database only stores the structured information and coding results of the cause analysis report, which is used to identify abnormal trends and common problems.

[0047] Step 1.4: Sort out the text data and coding results of the cause analysis report of high-level events in nuclear power plants, process the text information according to unified rules, and use a large language model for vectorization to build a high-level event trend coding knowledge base.

[0048] The cause analysis report text, coding results, coding logic, nuclear power basic knowledge base and other information are processed according to a unified preprocessing template, and then vectorized through a large language model to form a high-level event trend coding knowledge base. This vectorized knowledge base is different from the coding database, and it serves as the basis for automatic coding of the large model.

[0049] Step 2: Based on the high-level event trend coding knowledge base and large language model, an automatic coding model for the cause analysis report of high-level events in nuclear power plants is constructed.

[0050] The large language model is used to fine-tune the automatic encoding based on the trend encoding knowledge base to build an automatic encoding model. The existing encoding knowledge vector database is used as a training sample, and the prediction results are judged and fine-tuned based on human feedback reinforcement learning to ensure that the output results reach the target accuracy.

[0051] Step 3: Automatically encode and manually check the high-level event cause analysis report.

[0052] Step 3.1: Extract new high-level event cause analysis report information.

[0053] Extraction is carried out after the event cause analysis report is approved. It is necessary to extract unstructured information such as the description, consequences, cause analysis process and conclusion in the cause analysis report, and also to extract structured information such as the subject, unit, system equipment, time, etc. in the corresponding status report.

[0054] Step 3.2: Use the large language model to preprocess, segment, and vectorize the report information using a preset template to form a series of feature vectors, which are used as input to the large language model.

[0055] Step 3.3: Use the automatic coding model to automatically encode the high-level event cause analysis report, that is, generate it through vectorized data input combined with the coding knowledge base to obtain the coding result and the reason description for generating the coding.

[0056] Step 3.4: Manually check the coding and enter the final coding result into the coding database.

[0057] Step 4: Automatically identify abnormal trends and common problems in high-level events through coding results, and output the results.

[0058] Step 4.1: Set the criteria for determining abnormal trends and common problems.

[0059] Since the number of high-level events is relatively small, in order to ensure the identification of abnormal trends, the time measurement of the target code occurrence is needed as a feature. Here, the annual occurrence rate is used as the criterion for abnormal trend determination:

[0060] a. 1 point, the distance from the center is limited to be greater than K1 standard deviations;

[0061] b. K2 consecutive points, all increasing;

[0062] c. K3 consecutive points, with a distance from the center limit (above) greater than 2 standard deviations;

[0063] d. K4 consecutive points, with a distance from the center limit (above) greater than 1 standard deviation;

[0064] e. K5 consecutive points within 1 standard deviation from the center limit (above);

[0065] f. K6 consecutive points are above the center limit (excluding those falling on the center line).

[0066] The data points represent the annual incidence rate of a certain code. K1, K2, K3, K4, K5, and K6 are all natural numbers and can be flexibly adjusted according to actual conditions.

[0067] Step 4.2: Automatically identify abnormal trends and common problems of high-level events in nuclear power plants based on the judgment criteria and issue warnings, and output abnormal trend data, charts, trend overviews and other information.

[0068] Step 4.3: Manually check and confirm abnormal trends and common problems.

[0069] If the abnormal trend is confirmed as an unfavorable trend in the power plant's experience feedback area, management action should be taken to intervene. In this case, the corresponding abnormal trend evaluation conclusion needs to be generated and it is recommended that the management unit issue management actions.

[0070] Step 5: Track identified abnormal trends and common problems.

[0071] Step 5.1: Manually determine and include unusual trends that deserve attention into tracking.

[0072] Step 5.2: Extract the high-level event trend coding data of the new period and compare it with the existing trends in the abnormal trend tracking database.

[0073] Step 5.3: Output the trend tracking data for the current month.

[0074] Step 5.4: Manually determine whether the abnormal trend has been eliminated. If it has been eliminated, cancel the tracking; if it has not been eliminated, continue to include it in the tracking.

[0075] Here, abnormal trends can be tracked and determined on a monthly or annual basis. If the tracking code does not reappear for several consecutive months, or the interval between reappearances is much longer than the historical average interval, the corresponding abnormal trend can be untracked after a comprehensive evaluation confirms that the unfavorable trend has been eliminated.

[0076] The present invention is described in detail above with reference to the accompanying drawings and embodiments, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge of ordinary technicians in the field without departing from the purpose of the present invention. The contents not described in detail in the present invention can adopt the existing technology.

Claims

1. A method for intelligent trend analysis of high-level events in a nuclear power plant, characterized in that: The method comprises: Step 1: Sort out and manually code the cause analysis reports of high-level events to form a high-level event trend coding knowledge base; Step 2: Based on the high-level event trend coding knowledge base and large language model, an automatic coding model for the cause analysis report of high-level events in nuclear power plants is constructed; Step 3: Automatically encode and manually check the cause analysis report of high-level events; Step 4: Automatically identify abnormal trends and common problems in high-level events through coding results, and output the results; Step 5: Track identified abnormal trends and common problems.

2. According to claim 1, a method for intelligent trend analysis of high-level events in a nuclear power plant is characterized in that: The step one comprises: Step 1.1: Construct a high-level event classification and coding system; Step 1.2: Select the high-level event cause analysis report and manually code it according to the high-level event classification and coding system; Step 1.3: Enter the high-level event cause analysis report information and coding results into the coding database; Step 1.4: Sort out the text data and coding result information of the cause analysis report of high-level events in nuclear power plants, process the text information, and vectorize it using a large language model to build a high-level event trend coding knowledge base.

3. A method for intelligent trend analysis of high-level events in a nuclear power plant according to claim 2, characterized in that: In step 1.1, the high-level event codes are divided into equipment, human factors, technology and keyword codes.

4. A method for intelligent trend analysis of high-level events in a nuclear power plant according to claim 1, characterized in that: The step 2 includes: using a large language model to perform automatic encoding fine-tuning based on a trend encoding knowledge base to build an automatic encoding model.

5. The method for intelligent trend analysis of high-level events in a nuclear power plant according to claim 1, characterized in that: The step three comprises: Step 3.1: Extract new high-level event cause analysis report information; Step 3.2: Use the large language model to preprocess, segment, and vectorize the report information; Step 3.3: Use the automatic coding model to automatically code the high-level event cause analysis report, obtain the coding results, and generate the corresponding coding cause description; Step 3.4: Manually check the coding and enter the final coding result into the coding database.

6. A method for intelligent trend analysis of high-level events in a nuclear power plant according to claim 5, characterized in that: The cause analysis report information extracted in step 3.1 includes the unstructured information of the description, consequences, cause analysis process and conclusion in the cause analysis report, and the structured information of the subject, unit, system equipment and time in the corresponding status report.

7. A method for intelligent trend analysis of high-level events in a nuclear power plant according to claim 1, characterized in that: The fourth step comprises: Step 4.1: Set the criteria for determining abnormal trends and common problems; Step 4.2: Automatically identify abnormal trends and common problems of high-level events in nuclear power plants based on the judgment criteria and issue warnings, and output abnormal trend data, charts, trend overviews and other information; Step 4.3: Manually check and confirm abnormal trends and common problems.

8. A method for intelligent trend analysis of high-level events in a nuclear power plant according to claim 7, characterized in that: In step 4.3, if the abnormal trend is confirmed to be an unfavorable trend in the field of power plant experience feedback, management action should be taken to intervene. In this case, a corresponding abnormal trend evaluation conclusion needs to be generated and it is recommended that the management unit issue management actions.

9. The method for intelligent trend analysis of high-level events in a nuclear power plant according to claim 1, characterized in that: The step five comprises: Step 5.1: Manually determine and include the abnormal trends that deserve attention into tracking; Step 5.2: Extract the high-level event trend coding data of the new period and compare it with the existing trends in the abnormal trend tracking database; Step 5.3: Output the trend tracking data for the current month; Step 5.4: Manually determine whether the abnormal trend has been eliminated. If it has been eliminated, cancel the tracking; if it has not been eliminated, continue to include it in the tracking.

Images