CN107544337B - Method for intelligently classifying triggering data of loose part monitoring system - Google Patents
Method for intelligently classifying triggering data of loose part monitoring system Download PDFInfo
- Publication number
- CN107544337B CN107544337B CN201710843795.6A CN201710843795A CN107544337B CN 107544337 B CN107544337 B CN 107544337B CN 201710843795 A CN201710843795 A CN 201710843795A CN 107544337 B CN107544337 B CN 107544337B
- Authority
- CN
- China
- Prior art keywords
- data
- channel
- monitoring system
- loose part
- trigger
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Landscapes
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses a method for intelligently classifying loose part monitoring system trigger data, which comprises the following steps: acquiring original trigger data of a loose part monitoring system; obtaining data corresponding to each channel based on the original trigger data; carrying out waveform feature extraction on the data of each channel to obtain a feature vector of each channel; the method comprises the steps of establishing a single-channel classification model, obtaining a data classification result of each channel based on the single-channel classification model and the feature vector of each channel, intelligently distinguishing, counting and managing real loosening component impact signal data and false trigger data, and improving loosening component data analysis efficiency.
Description
Technical Field
The invention relates to the field of monitoring of loose parts of reactors and loops, in particular to a method for intelligently classifying trigger data of a loose part monitoring system.
Background
The reactor pressure vessel comprises an inner reactor component and a fuel assembly inside, wherein the inner reactor component mainly comprises a compaction assembly, a hanging basket assembly, a core lower supporting assembly and the like, and most parts and parts in the assemblies are connected through screws, pins and the like. Under the condition of long-term continuous operation of the pile, certain parts can be loosened and even fall off due to the impact of water flow and the influence of vibration caused by the flow, so that a loosening element is formed in a primary circuit. There is also the possibility of metal parts (called foreign parts) remaining in the primary loop system during construction, refuelling or maintenance. If the loose parts are not discovered and disposed in time, light persons cause huge economic loss, and heavy persons seriously affect the safe operation of the nuclear power station, and the consequences are not imagined. Therefore, all domestic nuclear power stations are provided with loose part monitoring systems for monitoring loose parts of the reactor and a primary circuit on line in real time.
The principle of the nuclear power plant reactor and a loop loosening component monitoring system is that when a reactor, a main pipeline, a main pump and a steam generator generate a moving part in coolant and a loosening part of equipment, the loosening part collides with the wall of the equipment or an internal component to generate a bending wave signal, the signal is transmitted along the wall of the equipment and the wall of the pipe, an accelerometer arranged on the wall of the equipment is used for detecting the sound wave signal, and the signal is processed and analyzed to judge whether the loosening component exists or not. Two problems arise due to the fact that interference with the reactor and the primary loop system and external operation of the unit (such as control rod drive mechanism movement, external operation or valve actuation, increased noise levels, system self-test, etc.) may cause the system to "false trigger", resulting in a large amount of trigger data:
firstly, real loose part impact data may be submerged in these "false trigger" data, affecting the timeliness and accuracy of loose part analysis;
second, system operation and maintenance personnel are heavy and mechanical in workload for triggering data analysis of the daily loose part system.
In summary, in the process of implementing the technical solution of the present invention, the inventors of the present application find that the above-mentioned technology has at least the following technical problems:
in the prior art, the existing loose part monitoring system has the technical problems of easy error triggering, poor analysis timeliness and accuracy and low triggering data analysis efficiency.
Disclosure of Invention
The invention provides a method for intelligently classifying loose part monitoring system trigger data, which is used for automatically eliminating data generated by interference or system self-detection so as to improve the analysis efficiency and the intelligent level of loose part system trigger signal data.
To achieve the above object, the present application provides a method for intelligently classifying trigger data of a loose part monitoring system, the method comprising:
acquiring original trigger data of a loose part monitoring system;
obtaining data corresponding to each channel based on the original trigger data;
carrying out waveform feature extraction on the data of each channel to obtain a feature vector of each channel;
and establishing a single-channel classification model, and obtaining a data classification result of each channel based on the single-channel classification model and the feature vector of each channel.
According to the invention, effective characteristic quantities are extracted aiming at massive loose part monitoring system trigger data, an intelligent classification model of the loose part monitoring system trigger data is established by using a machine learning method, real loose part impact signal data and false trigger data are intelligently distinguished, counted and managed, and the loose part data analysis efficiency is improved.
The method can realize the following functions:
(1) the loosening component monitoring system triggers intelligent classification of data;
(2) the loose part monitoring system triggers data classification statistics, display and management;
(3) the intelligent classification software can simultaneously start a plurality of program processes on the same host, and the processes do not influence each other.
Release member monitoring system trigger data a complete waveform data (release member monitoring system trigger data storage format is binary dat type file) is composed of 13 channel data, each channel contains 5000 data points. Therefore, the prediction of the entire waveform data depends on the prediction of each single-channel data. For this purpose, a machine learning method is adopted to learn different waveform data of a single channel, and a classification model is trained. And finally, combining the prediction results of each channel of the waveform data by using the model to form the final prediction result of the complete waveform data.
The loose part monitoring system triggers data characteristic quantity extraction:
when complex objects are studied, the original objects are typically mapped into a feature vector space by feature extraction. Each dimension of the feature vector corresponds to a feature of the original object. Mapping the original object to the feature vector space has two main benefits:
1) through feature extraction, the dimensionality of the original object can be effectively reduced. The loose part monitoring system triggers data that 5000 points are in a raw channel data, and the data dimension can be reduced to 46 dimensions through feature extraction. The low dimensional data enables machine learning algorithms to be trained more easily, resulting in higher quality models.
2) The feature extraction can conveniently blend preset conditions or data into a machine learning algorithm. Features are often definitions of key elements of a study object by a technician and play an important role in machine learning tasks. With the preset high quality features provided by the technician, the machine learning algorithm can obtain better results.
3) The model obtained by training the characteristic data has higher interpretability.
The intelligent model of the loose part monitoring system trigger data is established:
a model capable of classifying single-channel waveform data is generated by a machine learning algorithm. The invention adopts a Random Forest (Random Forest) algorithm to train a classification model. The random forest algorithm adopts an integrated learning technology to learn and generate a plurality of decision trees, and the prediction results of all the decision trees are fused to obtain a result. And converting the expert knowledge into a logic rule, and fusing the logic rule with a single-channel prediction result to finally form an intelligent classification model of the loose part monitoring system trigger data.
The performance index is shown in table 1:
TABLE 1
One or more technical solutions provided by the present application have at least the following technical effects or advantages:
by combining the knowledge base information of the technical skill in the art, 49 waveform characteristics based on time domain and frequency domain are designed and realized. The waveform characteristics have high interpretability and are of great significance for verifying and explaining the model effect;
the method comprises the following steps of effectively combining the characteristics of a waveform with a specific random forest algorithm in machine learning, expressing different characteristics in the waveform through different decision forms, and fusing technical knowledge base information to obtain a final classification model;
the model is subjected to software, a man-machine interaction interface is developed, and intelligent classification, data management and display of loose part trigger data are achieved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the embodiments of the invention and are incorporated in and constitute a part of this application, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention;
FIG. 1 shows a training process of a single-channel waveform recognition model of release part trigger data;
FIG. 2 the release unit triggers the data sorting process;
FIG. 3 is a flow of data sorting software.
Detailed Description
The invention provides a method for intelligently classifying loose part monitoring system trigger data, which is used for automatically eliminating data generated by interference or system self-detection so as to improve the analysis efficiency and the intelligent level of loose part system trigger signal data.
In order that the above objects, features and advantages of the present invention can be more clearly understood, a more particular description of the invention will be rendered by reference to the appended drawings. It should be noted that the embodiments and features of the embodiments of the present application may be combined with each other without conflicting with each other.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, however, the present invention may be practiced in other ways than those specifically described and thus the scope of the present invention is not limited by the specific embodiments disclosed below.
The whole algorithm training process is shown in fig. 1, and the model application process is shown in fig. 2.
Effective channel data extraction:
the loosening component monitoring system comprises six types of triggering data including a control rod impact signal, a typical component falling impact signal, a system self-checking signal, a pulse spike, water conservancy fluctuation and external interference. The data is labeled by taking the whole waveform data as a unit, for example, one loose part trigger data is composed of 13 channels, 5000 data points are arranged in each channel, only 1 channel of the waveform data labeled as water conservancy fluctuation is possible to be a water conservancy fluctuation signal, and the other 12 channels are all normal noise signals, so that the channel data which can really represent the waveform data category is called as effective channel data.
Feature extraction:
the loose part trigger data is mapped into the feature vector space by feature extraction. Each dimension of the feature vector corresponds to a feature of the original object. Mapping the original object to the feature vector space has three main benefits:
through feature extraction, the dimensionality of the original object can be effectively reduced. For example, there are 5000 points in a raw channel data, and the data dimension can be reduced to 49 dimensions by feature extraction. The low dimensional data enables machine learning algorithms to be trained more easily, resulting in higher quality models.
The feature extraction can conveniently integrate the knowledge base information of the technology in the field into a machine learning algorithm. Features are often definitions of key elements of study objects by those skilled in the art, and play an important role in machine learning tasks. With the high quality features provided by state of the art knowledge base information, machine learning algorithms can achieve better results.
The model obtained by training the characteristic data has higher interpretability.
And extracting event characteristic quantities such as peak amplitude, rise time, delay time, dominant frequency of a basic channel responding to a trigger interval, and peak amplitude, rise time, delay time, time difference and the like of a minimum time difference channel and a secondary time difference channel for the trigger signal of the loosening component based on expert knowledge. Aiming at the characteristic analysis of the waveform, some characteristic quantities which do not have the physical impact significance of the loose part but can effectively distinguish the waveform are extracted, because each characteristic respectively corresponds to a time domain characteristic and a frequency domain characteristic, and the quantity of all the characteristics is 49
Model training:
and generating a single-channel waveform data classification model capable of establishing loose part trigger data through a machine learning algorithm. And then fusing the prediction results of the single channels based on the information of the technical knowledge base in the field to form the final prediction result of the original waveform data.
The invention adopts a Random Forest (Random Forest) algorithm to train the model. And the random forest adopts an integrated learning technology to learn and generate a plurality of decision trees, and the prediction results of all the decision trees are fused to form a final result. The machine forest algorithm trains n decision trees together, and then the output result of the sample Instance is obtained by voting the results of the Instance of all the n decision trees. It is noted here that the features used by the random forest algorithm in training each decision tree are different (and may also be different in number). For example, in training the first decision tree, the top 10 features may be used, while in training the second decision tree, the 9 th to 15 th features may be used.
Through above-mentioned model, realized that the pine takes off the intelligent classification that the part triggered data single channel data, to the data of 13 passageways, combined expert's knowledge and judged the single channel data result, to not belonging to the above 6 type models, define this data type as other, finally realize the final classification of data.
Referring to fig. 3, the software implementation:
the waveform intelligent classification is an installation-free, cross-platform, network-free, desktop application with a Graphical User Interface (GUI). It provides a friendly interactive interface for users, which is used for assisting in the identification of large-batch waveforms and the real-time result statistics.
The application is based on an advanced desktop application construction technology, a single programming language and an open source technology framework are used, the popular WEB technology is combined, the difference among different operating systems is shielded, and one-time coding and multi-platform operation can be realized.
And after the user decompresses the application compression packet of the corresponding platform and the CPU architecture, double-clicking the executable program to open an application graphical interface. In this interface, the user needs to configure two parameters: a waveform data folder to be predicted and a prediction result output path.
The program automatically reads the waveform data folder to be predicted, displays all files and primary subfolder names under the folder, and counts the number of the files and the primary subfolder names. The user may also point the mouse to a file to see its creation time, modification time, and file size.
After a user clicks a start prediction button, the program verifies two input parameters, then the Java CLI is called, user input parameter information is transmitted, a built-in model file path is provided, and the prediction process is processed by the Java program.
After receiving the prediction instruction, the Java starts to perform waveform prediction and continuously generates a prediction result in an output path. The program starts a timer, polls the prediction result, and displays the waveform type, the number, the running state, the starting/ending time, the number of the identified waves and the average identification time consumption information in real time.
The method utilizes a machine learning method, extracts effective characteristic quantity through analysis of the waveform of trigger data, adopts a random forest algorithm, establishes an intelligent classification model of the trigger data of the loose part monitoring system through continuous training and combining expert knowledge, makes the model soft, effectively realizes intelligent distinguishing, counting and management of impact signal data of the loose part and false trigger data, and improves the efficiency of loose part data analysis.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the invention.
It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.
Claims (2)
1. A method of intelligent classification of loose part monitoring system trigger data, the method comprising:
acquiring original trigger data of a loose part monitoring system; the categories of the loose part monitoring system raw trigger data include: a control rod impact signal, a typical part falling impact signal, a system self-checking signal, a pulse spike signal, a water conservancy fluctuation signal and an external interference signal;
obtaining data corresponding to each channel based on the original trigger data;
carrying out waveform feature extraction on the data of each channel to obtain a feature vector of each channel; the specific process of extracting the waveform characteristics of the data of each channel and obtaining the characteristic vector of each channel is as follows: mapping each original channel data to a feature vector space through waveform feature extraction, wherein each dimension of a feature vector corresponds to one feature of each original channel data; extracting the peak amplitude, the rise time, the delay time and the dominant frequency of a basic channel responding to a trigger interval, and extracting the peak amplitude, the rise time, the delay time and the characteristic quantity of a time difference event of a minimum time difference channel and a secondary time difference channel for a loose part trigger signal;
establishing a single-channel classification model, and obtaining a data classification result of each channel based on the single-channel classification model and the feature vector of each channel; the specific process of establishing the single-channel classification model comprises the following steps: carrying out model training by adopting a random forest algorithm; the random forest algorithm is trained by adopting an integrated learning technology to generate a plurality of decision trees, and the characteristics used in training each decision tree are different;
and after the data classification result of each channel is obtained, combining the data classification results of each channel to form a prediction classification result of the original trigger data.
2. The method of claim 1, wherein the release member monitoring system trigger data is a waveform data consisting of 13 channels of data, each channel containing 5000 data points.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710843795.6A CN107544337B (en) | 2017-09-19 | 2017-09-19 | Method for intelligently classifying triggering data of loose part monitoring system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710843795.6A CN107544337B (en) | 2017-09-19 | 2017-09-19 | Method for intelligently classifying triggering data of loose part monitoring system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107544337A CN107544337A (en) | 2018-01-05 |
| CN107544337B true CN107544337B (en) | 2020-10-30 |
Family
ID=60963985
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710843795.6A Active CN107544337B (en) | 2017-09-19 | 2017-09-19 | Method for intelligently classifying triggering data of loose part monitoring system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107544337B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109543093A (en) * | 2018-09-29 | 2019-03-29 | 中国电子科技集团公司电子科学研究院 | A kind of integral intelligent information assistant's platform |
| CN109543894B (en) * | 2018-11-15 | 2020-11-24 | 深圳中广核工程设计有限公司 | System and method for predicting loose parts of nuclear power station in advance |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101718581A (en) * | 2009-11-13 | 2010-06-02 | 浙江大学 | Alarming method of nuclear power station loose part based on support vector machine |
| CN103093840A (en) * | 2013-01-25 | 2013-05-08 | 杭州电子科技大学 | Reactor loose part alarm method |
| CN103137224A (en) * | 2013-01-25 | 2013-06-05 | 杭州电子科技大学 | Nuclear power station loose part quality estimation method based on wavelet energy spectrum |
| CN104408303A (en) * | 2014-11-20 | 2015-03-11 | 西安交通大学 | Laser pulse mass spectrometry (LPMS) mass estimating method based on data matching |
| CN105702305A (en) * | 2016-02-03 | 2016-06-22 | 杭州电子科技大学 | Multi-target three-dimensional dynamic tracking method of loose parts of nuclear power plant |
Family Cites Families (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPS5764128A (en) * | 1980-10-08 | 1982-04-19 | Chugoku Electric Power Co Ltd:The | Loose parts monitor |
| JP2862876B2 (en) * | 1988-06-23 | 1999-03-03 | バブコツク日立株式会社 | Loose parts monitoring device |
| CN101685679B (en) * | 2008-09-27 | 2012-05-16 | 中国核动力研究设计院 | System for monitoring loosening part of nuclear reactor and coolant system |
| CN101750197B (en) * | 2008-12-11 | 2011-08-17 | 中国核动力研究设计院 | Loose part diagnosis method for nuclear power station |
| JP5129844B2 (en) * | 2010-09-07 | 2013-01-30 | 株式会社日本製鋼所 | Bolt tensioner and bolt tensioner adjusting device |
| CN103137225B (en) * | 2013-01-25 | 2015-09-30 | 杭州电子科技大学 | Based on the nuclear power station loose positioning parts method of wavelet transformation and Hilbert transform |
| JP2016057206A (en) * | 2014-09-10 | 2016-04-21 | 三菱重工業株式会社 | Loose parts monitoring device |
| CN104376881B (en) * | 2014-11-20 | 2016-10-26 | 西安交通大学 | Based on Hilbert transform and the positioning method for loosening member of nuclear power station of data screening |
| CN105225433B (en) * | 2015-08-22 | 2017-10-10 | 上海核工程研究设计院 | A kind of nuclear power station loose part monitor and alarm system and method |
| CN204886920U (en) * | 2015-08-22 | 2015-12-16 | 上海核工程研究设计院 | Not hard up part of nuclear power station signal conditioning circuit for monitoring system |
| CN105427904A (en) * | 2015-11-13 | 2016-03-23 | 陕西卫峰核电子有限公司 | Loose part positioning method based on nuclear power station |
| CN205982317U (en) * | 2016-06-22 | 2017-02-22 | 江苏核电有限公司 | Nuclear power station pine takes off part acceleration sensor installation fixing device |
| CN106874896B (en) * | 2017-03-31 | 2021-03-02 | 阳江核电有限公司 | Auxiliary learning method and system for signal feature identification of nuclear power station primary loop component loosening diagnosis system |
-
2017
- 2017-09-19 CN CN201710843795.6A patent/CN107544337B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101718581A (en) * | 2009-11-13 | 2010-06-02 | 浙江大学 | Alarming method of nuclear power station loose part based on support vector machine |
| CN103093840A (en) * | 2013-01-25 | 2013-05-08 | 杭州电子科技大学 | Reactor loose part alarm method |
| CN103137224A (en) * | 2013-01-25 | 2013-06-05 | 杭州电子科技大学 | Nuclear power station loose part quality estimation method based on wavelet energy spectrum |
| CN104408303A (en) * | 2014-11-20 | 2015-03-11 | 西安交通大学 | Laser pulse mass spectrometry (LPMS) mass estimating method based on data matching |
| CN105702305A (en) * | 2016-02-03 | 2016-06-22 | 杭州电子科技大学 | Multi-target three-dimensional dynamic tracking method of loose parts of nuclear power plant |
Non-Patent Citations (2)
| Title |
|---|
| 基于支持向量机的核电站松动件质量估计方法;郑华文;《浙江大学学报》;20100831;全文 * |
| 核电站松动部件的质量估计方法;何元峰;《知网》;20140215;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107544337A (en) | 2018-01-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108664269B (en) | A kind of feature attachment code peculiar smell detection method based on deep learning | |
| CN102130783B (en) | Intelligent alarm monitoring method of neural network | |
| CN105373472B (en) | A kind of method of testing and test system of the statistical accuracy based on database | |
| CN109462575A (en) | A kind of webshell detection method and device | |
| CN112311803B (en) | Rule base updating method and device, electronic equipment and readable storage medium | |
| CN107391598A (en) | One kind threatens information automatic generation method and system | |
| CN107544337B (en) | Method for intelligently classifying triggering data of loose part monitoring system | |
| CN109871686A (en) | Rogue program recognition methods and device based on icon representation and software action consistency analysis | |
| CN109800895A (en) | A method of based on augmented reality in the early warning of metering automation pipeline stall and maintenance | |
| CN105841793A (en) | Optical fiber vibration source identification method, device and system | |
| CN110175085A (en) | A kind of Hadoop system abnormal cause diagnostic method and device using map analysis | |
| CN109308411A (en) | The method and system of layered weighting software action defect based on artificial intelligence decision tree | |
| CN106370631B (en) | A kind of automatic assay of sepectrophotofluorometer and data acquisition and recording method | |
| CN111045902A (en) | Pressure testing method and device for server | |
| Xu et al. | Multi-Granularity Code Smell Detection using Deep Learning Method based on Abstract Syntax Tree. | |
| Tagaris et al. | Putting together wavelet-based scaleograms and convolutional neural networks for anomaly detection in nuclear reactors | |
| CN101930401B (en) | Detection object-based software vulnerability model detection method | |
| CN104331361B (en) | A kind of test device and method for white-box testing coverage rate calculation visualization | |
| CN117608889A (en) | Log semantic based anomaly detection method and related equipment | |
| CN112803587A (en) | Intelligent inspection method for state of automatic equipment based on diagnosis decision library | |
| JI et al. | Log Anomaly Detection Through GPT-2 for Large Scale Systems | |
| CN103701671A (en) | Method and device for detecting conflicts among businesses | |
| CN110309656A (en) | A kind of implicit type conversion safety detecting method | |
| CN110780660A (en) | Tobacco production industry control system fault diagnosis method based on production state | |
| CN113268419A (en) | Method, device, equipment and storage medium for generating test case optimization information |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |