CN113868948A - User-oriented dynamic threshold model training system and method - Google Patents
User-oriented dynamic threshold model training system and method Download PDFInfo
- Publication number
- CN113868948A CN113868948A CN202111140673.3A CN202111140673A CN113868948A CN 113868948 A CN113868948 A CN 113868948A CN 202111140673 A CN202111140673 A CN 202111140673A CN 113868948 A CN113868948 A CN 113868948A
- Authority
- CN
- China
- Prior art keywords
- model
- data
- module
- user
- dynamic threshold
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/957—Browsing optimisation, e.g. caching or content distillation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention particularly relates to a user-oriented webpage dynamic threshold model training system, which comprises a system interface, a data import module, a data selection module and a model training module, wherein the data import module, the data selection module and the model training module are connected with the system interface; the data import module is used for importing the measuring point data of different data sources, carrying out homologous processing on the measuring point data of the different data sources, generating original data and inputting the original data to the data selection module; the data selection module is used for selecting data required by model training from the original data, generating a normal sample set and inputting the normal sample set to the model training module; the model training module is used for generating a verification sample set and a health sample matrix, and performing model training after a user selects the health sample matrix on a system interface, and displaying the trained model to the user in a multi-group array mode. The dynamic threshold model training system provided by the invention can meet the requirement that a user completes the training of the dynamic threshold model through simple configuration operation, and realizes more effective supervision on equipment by using the dynamic threshold model.
Description
Technical Field
The invention relates to the technical field of dynamic threshold model training, in particular to a user-oriented dynamic threshold model training system and method.
Background
In the industrial field, parameters representing the running state of equipment, such as vibration, temperature, pressure, flow, current, voltage, liquid level and the like, are collected, and the online supervision of the equipment state is realized by setting a fixed alarm threshold value, so that the method is an effective means for avoiding equipment failure and relieving accident consequences. At present, the online monitoring mode of the equipment mainly sets an alarm threshold value for a single parameter, and when real-time data of a measuring point reaches the alarm threshold value, an alarm signal is sent out. With the increase of the complexity of the equipment, the disadvantage of the supervision mode gradually appears, and the salient phenomenon is as follows: the method can not respond to the changes of the running states such as equipment shutdown, unit power reduction and the like, and false alarm or alarm missing can be easily generated when the running state of the equipment changes frequently; the degradation trend of the equipment cannot be sensed, alarm information is easy to lag, and timely and effective decision making is not facilitated; the internal correlation between equipment parameters cannot be reflected, and the alarm information is one-sided and isolated.
Aiming at the problems, a dynamic threshold algorithm is gradually introduced in the industrial field to realize multi-parameter combined alarm, which is called as dynamic threshold alarm. When the dynamic threshold algorithm is relatively uniform, the dynamic threshold alarm effectiveness depends mainly on the dynamic threshold model. The dynamic threshold model is a group of characteristic values obtained by training original data of a plurality of equipment parameters with certain correlation by using a specific dynamic threshold algorithm, and after the training of the dynamic threshold model is finished, the real-time parameters of the equipment are led into the model, so that the dynamic threshold alarm can be realized. However, an important factor restricting popularization and application of the dynamic threshold value alarm is that training of a dynamic threshold value model often requires that users have certain algorithm and programming capabilities, and equipment engineers usually do not have the capabilities, so that the dynamic threshold value alarm cannot be effectively used to carry out more effective supervision on responsible equipment.
Disclosure of Invention
Based on this, it is necessary to provide a user-oriented dynamic threshold model training system and method for solving the problem that the existing dynamic threshold model training limits the equipment engineers who do not have software development ability to use dynamic threshold alarms to more effectively supervise the equipment in charge, so that the user can implement the work of constructing, training and the like of the dynamic threshold model through simple configuration operation, thereby implementing more effective supervision of the equipment by using the dynamic threshold model.
In order to achieve the above purpose, the invention provides the following technical scheme:
a user-oriented webpage dynamic threshold model training system comprises a system interface, and a data import module, a data selection module and a model training module which are connected with the system interface;
the data import module is used for importing the measuring point data of different data sources, carrying out homologous processing on the measuring point data of the different data sources, generating original data and inputting the original data to the data selection module;
the data selection module is used for selecting data required by model training from the original data, generating a normal sample set and inputting the normal sample set to the model training module;
the model training module is used for generating a verification sample set and a health sample matrix, and performing model training after a user selects the health sample matrix on a system interface, and displaying the trained model to the user in a multi-group array mode.
Furthermore, the data required by model training is data of not less than one overhaul period in a time range, the normal sample set is obtained by removing abnormal operation conditions and selecting normal operation conditions from original data by a user, the healthy sample matrix is obtained from the normal sample set through a sample selection algorithm, and the verification sample set is obtained by performing down-sampling and filtering sample combination on the remaining samples after the healthy sample matrix is removed from the normal sample set.
Furthermore, a three-point sample extraction algorithm, a five-point sample extraction algorithm and an equal-interval sampling algorithm are integrated on the model training module.
Furthermore, the user-oriented webpage dynamic threshold model training system also comprises a model basic information configuration module and an associated measuring point selection module which are connected with a system interface;
the model basic information configuration module is used for inputting the model basic information to the data import module after a user configures the model basic information on a system interface;
the associated measuring point selection module is used for inputting associated measuring point information to the data import module after a user selects a measuring point required to be associated for model training on a system interface;
the data import module is used for importing the measuring point data of the associated measuring point in the configuration time range and the sampling interval from the measuring point database after a user configures the time range and the sampling interval on the system interface, and carrying out homologous processing on the measuring point data of different data sources to generate original data and inputting the original data to the data selection module; the configuration time range is not less than one overhaul period.
Further, the basic information of the model configured by the user on the system interface comprises the name, the code, the version information, the object information and the maximum vector number of the health sample matrix of the model filled or modified by the user on the system interface; the object information comprises an object type, an object name and an object code; the object types include devices and systems.
Further, for a measuring point database supporting an external interface, the data import module imports measuring point data of the associated measuring points in a configuration time range and a sampling interval from the measuring point database through an interface of the measuring point database; for the measuring point database which does not support external interfaces, the data import module imports measuring point data of associated measuring points in a configuration time range and a sampling interval from the measuring point database by deploying to the measuring point database.
Further, the associated measuring point selection module is used for giving a prompt in the form of a dialog box when the user selects the measuring point which is also used for the user to select the association required by the model training in the system interface, the prompt comprises a measuring point selection mode and the association degree of the measuring point, the measuring point selection mode is given by analyzing the actual mechanism of the object and the association degree of the measuring point, and the association degree of the measuring point is given by a similarity algorithm.
Further, the user-oriented webpage dynamic threshold model training system further comprises a model verification module connected with a system interface, and a similarity matching algorithm is integrated on the model verification module;
the data selection module is also used for inputting the filtered data to the model verification module;
the model training module is also used for inputting the trained model, the health sample matrix and the verification sample set into the model verification module;
the model verification module is used for generating a part of verification sample set and inputting the part of verification sample set to the model evaluation module after a user manually selects data from the verification sample set on a system interface; and after the user selects the similarity matching algorithm on the system interface, verifying the prediction effect of the selected verification sample set in the trained model, and displaying the verification result to the user in an interface form.
Furthermore, the verification result is an actual value, a predicted value and a residual value of the model verification sample set, and the residual value is a residual between the predicted value and the actual value, so that the precision of the training model can be evaluated; the similarity matching algorithm comprises a Manhattan distance, a Euclidean distance and a cosine similarity.
Further, the model verification module is further configured to determine an alarm range of the dynamic threshold by automatically or manually generating upper and lower limits of the dynamic threshold.
Further, the upper and lower limits of the dynamic threshold are automatically calculated based on three sigma law. Further, the upper and lower limits of the dynamic threshold are secondarily corrected by an experienced expert based on self experience in combination with the analysis result of the mathematical model.
Further, the user-oriented webpage dynamic threshold model training system further comprises a model evaluation module connected with a system interface;
the model verification module is also used for inputting the verification sample set and the verified model into the model evaluation module;
the model evaluation module is used for calculating an error value of the verification sample set in the model which is verified, displaying the error value to a user in an interface mode, and evaluating the accuracy of the model which is verified according to the error value by the user.
Further, the error values include an interpretable variance, a deterministic coefficient, a mean absolute error, a root mean square error, a maximum error, and a mean relative error, for evaluating model accuracy.
Furthermore, the user-oriented webpage dynamic threshold model training system also comprises a model publishing module connected with a system interface;
the model evaluation module is also used for inputting the evaluated model to the model publishing module;
the model issuing module is used for issuing the evaluated model after a user clicks a 'model issuing' button on a system interface, enabling the evaluated model to be online and used for sensing the state of equipment in real time, and achieving a dynamic threshold value alarming function.
Further, the data selection module is also used for counting data selection results and inputting the data selection statistics results to the model publishing module, and the model publishing module is also used for displaying the data selection statistics results to a user in an interface form, such as the number of filtering samples and the number of adding samples.
Further, the data selection module comprises an automatic data filtering module, a manual data filtering module and a manual data selection module;
the data import module is used for inputting original data to the data automatic filtering module;
the data automatic filtering module is used for automatically filtering the original data meeting the filtering condition after a user sets the filtering condition according to experience on a system interface, the automatically filtered data is input to the data manual filtering module, and the automatically filtered data is input to the model verification module;
the data manual filtering module is used for manually filtering the automatically filtered data according to a graph zooming function provided by a webpage by a user on a system interface, inputting the manually filtered data to the data manual selection module, and inputting the manually filtered data to the model verification module;
the data manual selection module is used for generating a normal sample set by the selected data and inputting the normal sample set into the model training module after a user manually selects and determines data required by model training from the manually filtered data by setting a time range and a sampling interval on a system interface, and inputting the unselected data into the model verification module.
Furthermore, the filtering conditions comprise working condition filtering conditions, threshold filtering conditions and null value filtering conditions, the automatic data filtering module automatically rejects the start-stop working condition data sets through the working condition filtering, rejects the super-threshold and low-threshold data sets through the threshold filtering, and rejects the missing value data sets through the null value filtering.
Further, the data automatic filtering module displays the automatically filtered data by using a red square frame; the data manual filtering module displays manually filtered data by using a red square frame; the data manual selection module displays the unselected data by a red square box and displays the selected data by a green square box.
Further, the data selection result comprises a data automatic filtering result, a data manual filtering result and a data manual selection result; the data automatic filtering module is also used for counting the automatic filtering result of the original data and inputting the automatic filtering result of the original data to the model publishing module; the data manual filtering module is also used for counting the manual filtering results of the original data and inputting the manual filtering results of the original data to the model issuing module; the data manual selection module is also used for counting the manual selection result of the original data and inputting the manual selection result of the original data to the model publishing module.
A user-oriented webpage dynamic threshold model training method comprises the following steps:
1. importing measuring point data of different data sources, and carrying out homologous processing on the measuring point data of the different data sources to generate original data of not less than one overhaul period;
2. selecting data required by model training from the original data, and generating a normal sample set through a manual selection module;
3. and selecting a health sample matrix algorithm, training the state monitoring model by using the health sample matrix algorithm, and displaying the trained model to a user in a mode of a plurality of groups of data points.
Further, after step 3, the user-oriented web page dynamic threshold model training method further includes the following steps:
4. and verifying the prediction effect of the selected verification sample set in the trained model by combining the selected verification sample set with the selected similarity matching algorithm, and visually displaying the verification result to the user in an interface form.
Further, step 4, also includes the following steps: and the alarm range of the dynamic threshold is determined by automatically or manually generating the upper limit and the lower limit of the dynamic threshold.
Further, the upper and lower limits of the dynamic threshold are automatically calculated based on three-sigma law, and further, the upper and lower limits of the dynamic threshold are secondarily corrected by an experienced expert based on self experience and combined with the analysis result of the mathematical model.
Further, after step 4, the user-oriented webpage dynamic threshold model training method further includes the following steps:
5. and calculating an error value of the model verification sample set in the model after verification is completed, and displaying the error value to a user in an interface mode.
Further, after step 5, the user-oriented web page dynamic threshold model training method further includes the following steps:
6. and releasing the model after the model evaluation is completed, and enabling the model after the model evaluation to be online and used for sensing the state of the equipment in real time to realize the dynamic threshold value alarm function.
The invention has the beneficial technical effects that:
the invention provides a user-oriented webpage-side dynamic threshold model training system and method, which realize links such as model basic information configuration, associated measuring point screening, measuring point data import, data automatic filtering, data manual filtering, model construction, model training, model verification, model release and the like on a webpage side by modularizing steps and algorithms of dynamic threshold model training and using a software programming mode, and can implement model training, verification and release work only by qualitatively knowing equipment related knowledge and equipment state data without using personnel to have programming and algorithm bases.
The user-oriented webpage dynamic threshold model training system and method provided by the invention are internally provided with the core steps and common algorithms of dynamic threshold model training, conveniently and efficiently realize a whole set of processes of construction, training, verification, release and the like of the dynamic threshold model, ensure users without algorithms and programming capability, and can also use the system to complete the training of the dynamic threshold model based on the experience and knowledge of the users on the performance and state of equipment.
Drawings
FIG. 1 is a flowchart of a user-oriented web page dynamic threshold model training method of the present invention;
FIG. 2 is a schematic structural diagram of a user-oriented web page dynamic threshold model training system according to the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and examples.
Referring to fig. 1-2, the invention provides a user-oriented webpage dynamic threshold model training system, which comprises a system interface, and a model basic information configuration module, an associated measuring point selection module, a data import module, a data selection module, a model training module, a model verification module, a model evaluation module and a model release module which are connected with the system interface, wherein the data selection module comprises an automatic data filtering module, a manual data filtering module and a manual data selecting module.
The model basic information configuration module is used for inputting the model basic information to the data import module after a user configures the model basic information on the system interface. The basic information of the model configured on the system interface by the user comprises the name, the code, the version information, the object information and the maximum vector number of the health sample matrix of the model filled or modified on the system interface by the user; the object information of the model comprises an object type of the model, an object name of the model and an object code of the model; the object types of the model include devices and systems. The name and code of the model are used for distinguishing each model by a user, the version information of the model is convenient for the user to control the version of the model, the object information of the model is used for the user to clearly detect the related content of the object, and the maximum vector number of the health sample matrix is not more than 500 groups and is used for controlling the maximum sample number.
The relevant measuring point selection module is used for giving a prompt in a dialog box form when a user selects a measuring point required to be relevant for model training according to self experience on a system interface, wherein the prompt comprises a measuring point selection mode and a measuring point relevance degree, the measuring point selection mode is given by analyzing the actual mechanism of an object and the relevance degree between the measuring points, the relevance degree of the measuring points is given by a similarity algorithm, and a user monitoring model construction mode can be standardized under the relevant measuring point selection module; and the data import module is used for inputting the relevant measuring point information to the data import module after the user selects the relevant measuring point required by the model training according to the experience of the user and the prompt given by the auxiliary relevant measuring point selection module on the system interface.
The data import module is used for importing the measuring point data of the associated measuring point in the configuration time range and the sampling interval from the measuring point database after a user configures the time range and the sampling interval on the system interface, carrying out homologous processing on the measuring point data of different data sources, generating original data and inputting the generated original data into the automatic data filtering module, wherein the selected time range is not less than one overhaul period. The sampling interval is set mainly to ensure that the total number of the original data does not exceed 10000 groups, and the sampling interval can be automatically calculated by the system based on the criterion.
For the measuring point database supporting an external interface, the data import module has the characteristics of low delay and high concurrency through the interface with the measuring point database, and the measuring point data of the associated measuring points in the configuration time range and the sampling interval are imported from the measuring point database; for the measuring point database which does not support external interfaces, the data import module imports measuring point data of associated measuring points in a configuration time range and a sampling interval from the measuring point database by deploying to the measuring point database. The multi-source data imported in the data import module are strictly aligned through the time stamps.
The data automatic filtering module is used for automatically filtering original data meeting filtering conditions after a user sets filtering conditions such as working condition filtering, threshold values and null values according to experience on a system interface, and comprises the steps of automatically rejecting start-stop working condition data sets through the working condition filtering conditions, automatically rejecting super-threshold data sets and low-threshold data sets through the threshold value filtering conditions, and automatically rejecting missing value data sets through the null value filtering conditions; and the module is used for displaying the automatically filtered data by using a red square frame, inputting the automatically filtered data to the data manual filtering module, and inputting the automatically filtered data to the model verification module.
The data manual filtering module is used for manually filtering the automatically filtered data by a user according to a graph zooming function provided by a webpage on a system interface, displaying the manually filtered data by a red square frame, inputting the manually filtered data to the data manual selecting module, and inputting the manually filtered data to the model verifying module.
The data manual selection module is used for manually selecting and determining data required by model training from manually filtered data by setting a time range and a sampling interval on a system interface by a user, displaying the selected data by using a green square frame, displaying the unselected data by using a red square frame, generating a normal sample set from the selected data, inputting the normal sample set to the model training module, and inputting the unselected data to the model verification module.
The model training module is integrated with a three-point sample extraction algorithm, a five-point sample extraction algorithm and an equidistant sampling algorithm, wherein the three-point sample extraction algorithm and the five-point sample extraction algorithm are used for extracting characteristic (mean value, maximum value, minimum value, upper quantile and lower quantile) data in a designated window to generate a healthy sample matrix of no more than 500 groups of samples for constructing a monitoring model, the model training module is used for a user to select a state matrix algorithm on a system interface in the system, a 'generation model' button is clicked to generate a state monitoring model, remaining samples after the construction of the state matrix are removed from a normal sample set are combined with filtered samples to generate a verification sample set, and the trained model and the verification sample set are input to the model verification module.
And the model verification module integrates similarity matching algorithms such as Manhattan distance, Euclidean distance, cosine similarity and the like, the similarity of the verification sample set and the health state matrix is calculated to distribute weight, the distributed weight is larger when the similarity of the sample is higher, and the estimated value calculated by the weight and the health state matrix is the state value of the monitoring equipment in the health state and is compared with the actual verification set.
The model verification module is used for inputting a verification sample set generated under the model training module to the model evaluation module after a user selects a piece of data from data filtered from original data in a system interface; the measuring point characteristics mainly indicate that information such as the distribution range of measuring points and the standard deviation of the measuring points is used for verifying the prediction effect of a verification sample set in a trained model after a user selects a similarity matching algorithm on a system interface, and a verification result is displayed to the user in an interface form, wherein the verification result is an actual value, a predicted value and a residual error value of the model verification sample set, and the residual error value is a residual error between the predicted value and the actual value, so that the precision of the trained model can be evaluated; the model evaluation module is used for inputting the verified model to the model evaluation module; various indicators are introduced into the model evaluation module including, but not limited to, the interpretable variance, the mean absolute error, the maximum error, the mean relative error, the deterministic coefficient, and the root mean square error. Wherein the interpretable variance and the determination coefficient are between 0.8 and 1 to indicate that the algorithm identification effect is excellent, between 0.5 and 0.8 to indicate that the algorithm identification effect is good, and when the interpretable variance and the determination coefficient are less than 0.5 to indicate that the algorithm identification effect is general; the evaluation parameters such as the average absolute error, the root mean square error, the maximum error, the average relative error and the like need to be given by an engineer according to the characteristics of the measuring points; the upper limit and the lower limit of the dynamic threshold can be generated under the function page in an automatic or manual mode to determine the alarm range of the dynamic threshold. For the upper and lower limits of the dynamic threshold, the model verification module defaults to automatically calculate based on the sigma rule, and an experienced expert can perform secondary correction on the upper and lower limits of the dynamic threshold based on own experience and combined with the analysis result of the mathematical model.
The model evaluation module is used for calculating an error value of the verification sample set in the model which completes verification, the error value is displayed to a user in an interface mode, the user evaluates the accuracy of the model which completes verification according to the error value, and the error value comprises an interpretable variance, a determination coefficient, an average absolute error, a root mean square error, a maximum error and an average relative error; and the model releasing module is used for inputting the evaluated model to the model releasing module.
The model publishing module is used for displaying data selection statistical results such as the number of filtering samples and the number of adding samples to a user in an interface mode; and after clicking a 'model issuing' button on a system interface, a user issues the evaluated model, and the evaluated model is on line and used for sensing the state of the equipment in real time, so that the function of dynamic threshold value alarm is realized.
A user-oriented webpage dynamic threshold model training method comprises the following steps:
1. after a user configures the model basic information on the system interface, the model basic information configuration module sends the model basic information to the data import module.
2. When a user selects a measuring point required to be associated by model training on a system interface according to own experience, the associated measuring point selection module gives a prompt in a dialog box mode; after a user selects the measuring points required to be associated for model training on a system interface according to own experience and prompts given by the associated measuring point selection module, the associated measuring point selection module inputs associated measuring point information to the data import module.
3. After a user configures a time range and a sampling interval of a measuring point on a system interface, a data import module imports measuring point data of a related measuring point in the configured time range and the sampling interval from a measuring point database, performs homologous processing on the measuring point data of the related measuring point in the configured time range and the sampling interval, generates original data and inputs the original data to an automatic data filtering module.
4. A user sets filtering conditions such as working condition filtering, threshold values and null values on a system interface according to experience, the automatic data filtering module performs automatic filtering processing on original data meeting the filtering conditions, red square frames are used for displaying the automatically filtered data, the automatically filtered data are input to the manual data filtering module, and the automatically filtered data are input to the model verification module.
5. After a user selects the automatically filtered data section by section and point by point on a system interface by using an image zooming function provided by a webpage, the manually filtered data is manually filtered by the manually filtered data filtering module, the manually filtered data is displayed by a red square frame, the manually filtered data is input to the manually filtered data selecting module, and the manually filtered data is input to the model verifying module.
6. After a user sets a time range and a sampling interval on a system interface, the data manual selection module automatically selects manually filtered data in the time range and the sampling interval, the selected data are displayed by green square frames, the unselected data are displayed by red square frames, the selected data are generated into a normal sample set and input to the model training module, and the unselected data are input to the model verification module.
7. The method comprises the steps that a user checks a state matrix generation algorithm on a system interface, clicks a 'model generation' button, generates a health state matrix, generates a state monitoring model and a verification sample set, displays a model which is trained on the system interface in a multi-group data point mode, and inputs the trained model and the verification sample set to a model verification module.
8. After a user manually selects a piece of data from the filtered data on a system interface by using an image zooming function provided by a webpage, the model verification module generates a partial verification sample set and inputs the partial verification sample set to the model evaluation module; after a user selects a similarity matching algorithm on a system interface, a model verification module verifies the prediction effect of the selected verification sample set in the trained model, displays a verification result to the user in an interface mode, and inputs the verified model to a model evaluation module;
meanwhile, the model verification module can also generate the upper limit and the lower limit of the dynamic threshold value in an automatic or manual mode to determine the alarm range of the dynamic threshold value. Specifically, for the upper and lower limits of the dynamic threshold, the model verification module defaults to automatically calculate based on sigma rule, and an experienced expert can perform secondary correction on the upper and lower limits of the dynamic threshold based on own experience and combined with the analysis result of the mathematical model.
9. The model evaluation module calculates error values such as root mean square error, square absolute error, mean square error and the like of the model verification sample set in the model which is verified, displays the error values to a user in an interface mode, and inputs the model which is evaluated to the model release module.
10. After a user clicks a 'model issuing' button on a system interface, a model issuing module issues a model for completing model evaluation, and the model for completing model evaluation is on-line and used for sensing the state of equipment in real time, so that a dynamic threshold value alarming function is realized.
The method can be applied to scenes with monitoring, modeling and simulation requirements and without software programming skills of engineers, and almost comprises all fields of industrial manufacturing industry.
Typical application scenarios are: a certain power plant engineer hopes to carry out more in-depth monitoring and analysis on a system in charge of the engineer, the engineer quickly draws a flow chart which is basically consistent with a design drawing or a power station control system on a webpage end by comparing a system process flow chart through the platform, and builds a dynamic threshold model through the access of measuring points and verifies and releases the dynamic threshold model to achieve the purpose of monitoring the system state. Further, engineers can use the calculation and analysis functions to carry out more in-depth analysis work on the dynamic threshold change condition of each parameter under the monitored equipment.
The above-mentioned embodiments only express several embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.
Claims (15)
1. A user-oriented webpage dynamic threshold model training system is characterized by comprising a system interface, a historical data import module, a data selection module and a model training module, wherein the historical data import module, the data selection module and the model training module are connected with the system interface;
the data import module is used for importing the measuring point data of different data sources, carrying out homologous processing on the measuring point data of the different data sources, generating original data and inputting the original data to the data selection module;
the data selection module is used for selecting data required by model training from the original data, generating a normal sample set and inputting the normal sample set to the model training module;
the model training module is used for generating a verification sample set and a health sample matrix, and performing model training after a user selects the health sample matrix on a system interface, and displaying the trained model to the user in a multi-group array mode.
2. The user-oriented web page dynamic threshold model training system according to claim 1, wherein the data required for model training is data of not less than one overhaul period in a time range, the normal sample set is obtained by removing abnormal operation conditions and selecting normal operation conditions from original data by a user, the healthy sample matrix is obtained from the normal sample set by a sample selection algorithm, and the verification sample set is obtained by down-sampling and combining filtered samples of remaining samples after removing the healthy sample matrix from the normal sample set.
3. The user-oriented web-side dynamic threshold value model training system according to claim 1, wherein a three-point sample extraction algorithm, a five-point sample extraction algorithm and an equal-interval sampling algorithm are integrated on the model training module.
4. The user-oriented webpage-side dynamic threshold model training system according to claim 1, further comprising a model basic information configuration module and an associated measure point selection module connected with a system interface;
the model basic information configuration module is used for inputting the model basic information to the data import module after a user configures the model basic information on a system interface;
the associated measuring point selection module is used for inputting associated measuring point information to the data import module after a user selects a measuring point required to be associated for model training on a system interface;
the data import module is used for importing the measuring point data of the associated measuring point in the configuration time range and the sampling interval from the measuring point database after a user configures the time range and the sampling interval on the system interface, and carrying out homologous processing on the measuring point data of different data sources to generate original data and inputting the original data to the data selection module; the configuration time range is not less than one overhaul period.
5. The user-oriented webpage-side dynamic threshold value model training system of claim 4, wherein the associated measuring point selection module is further used for giving a prompt in a dialog box form when the user selects a measuring point which is also used for the user to select association required for model training on the system interface, the prompt comprises a measuring point selection mode and a measuring point association degree, the measuring point selection mode is given by analyzing the association degree between the actual mechanism of the object and the measuring point, and the measuring point association degree is given by a similarity algorithm.
6. The user-oriented webpage-side dynamic threshold model training system according to any one of claims 1 to 5, further comprising a model verification module connected with a system interface, wherein a similarity matching algorithm is integrated on the model verification module;
the data selection module is also used for inputting the filtered data to the model verification module;
the model training module is also used for inputting the trained model, the health sample matrix and the verification sample set into the model verification module;
the model verification module is used for generating a part of verification sample set and inputting the part of verification sample set to the model evaluation module after a user manually selects data from the verification sample set on a system interface; and after the user selects the similarity matching algorithm on the system interface, verifying the prediction effect of the selected verification sample set in the trained model, and displaying the verification result to the user in an interface form.
7. The user-oriented webpage-side dynamic threshold model training system of claim 6, wherein the model verification module is further configured to generate upper and lower limits of the dynamic threshold to determine an alarm range of the dynamic threshold, the upper and lower limits of the dynamic threshold are automatically calculated based on three-sigma law, and are secondarily corrected by an experienced expert based on own experience and a result of the mathematical model analysis.
8. The user-oriented web-site dynamic threshold model training system of claim 6, further comprising a model evaluation module interfaced with the system;
the model verification module is also used for inputting the verification sample set and the verified model into the model evaluation module;
the model evaluation module is used for calculating an error value of the verification sample set in the model which is verified, displaying the error value to a user in an interface mode, and evaluating the accuracy of the model which is verified according to the error value by the user.
9. The user-oriented web-site dynamic threshold model training system of claim 8, further comprising a model publishing module interfaced with the system;
the model evaluation module is also used for inputting the evaluated model to the model publishing module;
the model issuing module is used for issuing the evaluated model after a user clicks a 'model issuing' button on a system interface, enabling the evaluated model to be online and used for sensing the state of equipment in real time, and achieving a dynamic threshold value alarming function.
10. The user-oriented web-side dynamic threshold model training system of claim 8, wherein the data selection module is further configured to count data selection statistics and input the data selection statistics to the model publishing module, and the model publishing module is further configured to interface data selection statistics to the user, such as filter sample number and add sample number.
11. The user-oriented webpage-side dynamic threshold model training system of claim 8, wherein the data selection module comprises an automatic data filtering module, a manual data filtering module and a manual data selection module;
the data import module is used for inputting original data to the data automatic filtering module;
the data automatic filtering module is used for automatically filtering the original data meeting the filtering condition after a user sets the filtering condition according to experience on a system interface, the automatically filtered data is input to the data manual filtering module, and the automatically filtered data is input to the model verification module;
the data manual filtering module is used for manually filtering the automatically filtered data according to a graph zooming function provided by a webpage by a user on a system interface, inputting the manually filtered data to the data manual selection module, and inputting the manually filtered data to the model verification module;
the data manual selection module is used for generating a normal sample set by the selected data and inputting the normal sample set into the model training module after a user manually selects and determines data required by model training from the manually filtered data by setting a time range and a sampling interval on a system interface, and inputting the unselected data into the model verification module.
12. A user-oriented webpage dynamic threshold model training method comprises the following steps:
step 1, importing measuring point data of different data sources, and performing homologous processing on the measuring point data of the different data sources to generate original data of not less than one overhaul period;
step 2, selecting data required by model training from the original data, and generating a normal sample set through a manual selection module;
and 3, selecting a health sample matrix algorithm, training the state monitoring model by using the health sample matrix algorithm, and displaying the trained model to a user in a mode of a plurality of groups of data points.
13. The system for training the dynamic threshold value model of the user-oriented webpage end according to claim 12, wherein after the step 3, the method for training the dynamic threshold value model of the user-oriented webpage end further comprises the following steps:
and 4, verifying the prediction effect of the selected verification sample set in the trained model by combining the selected verification sample set and the selected similarity matching algorithm, and visually displaying the verification result to the user in an interface form.
14. The system for training the dynamic threshold value model of the user-oriented webpage end according to claim 12, wherein after the step 4, the method for training the dynamic threshold value model of the user-oriented webpage end further comprises the following steps:
and 5, calculating an error value of the model verification sample set in the model which is verified, and displaying the error value to a user in an interface mode.
15. The system for training the dynamic threshold value model of the user-oriented webpage end according to claim 12, wherein after the step 5, the method for training the dynamic threshold value model of the user-oriented webpage end further comprises the following steps:
and 6, releasing the model after the model evaluation is finished, enabling the model after the model evaluation to be online and used for sensing the equipment state in real time, and realizing a dynamic threshold value alarm function.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111140673.3A CN113868948A (en) | 2021-09-28 | 2021-09-28 | User-oriented dynamic threshold model training system and method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111140673.3A CN113868948A (en) | 2021-09-28 | 2021-09-28 | User-oriented dynamic threshold model training system and method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113868948A true CN113868948A (en) | 2021-12-31 |
Family
ID=78991678
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111140673.3A Pending CN113868948A (en) | 2021-09-28 | 2021-09-28 | User-oriented dynamic threshold model training system and method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113868948A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114565638A (en) * | 2022-01-25 | 2022-05-31 | 上海安维尔信息科技股份有限公司 | Multi-target tracking method and system based on tracking chain |
CN116628564A (en) * | 2023-04-20 | 2023-08-22 | 上海宇佑船舶科技有限公司 | Model training method and system for detecting generator state |
-
2021
- 2021-09-28 CN CN202111140673.3A patent/CN113868948A/en active Pending
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114565638A (en) * | 2022-01-25 | 2022-05-31 | 上海安维尔信息科技股份有限公司 | Multi-target tracking method and system based on tracking chain |
CN116628564A (en) * | 2023-04-20 | 2023-08-22 | 上海宇佑船舶科技有限公司 | Model training method and system for detecting generator state |
CN116628564B (en) * | 2023-04-20 | 2024-03-12 | 上海宇佑船舶科技有限公司 | Model training method and system for detecting generator state |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2022252505A1 (en) | Device state monitoring method based on multi-index cluster analysis | |
CN108375715B (en) | Power distribution network line fault risk day prediction method and system | |
CN108199795B (en) | A kind of monitoring method and device of equipment state | |
CN104573850A (en) | Method for evaluating state of thermal power plant equipment | |
CN115201608A (en) | Power plant equipment operation parameter monitoring method based on neural network | |
CN113868948A (en) | User-oriented dynamic threshold model training system and method | |
CN113036913B (en) | Method and device for monitoring state of comprehensive energy equipment | |
CN117176560B (en) | Monitoring equipment supervision system and method based on Internet of things | |
CN105654229A (en) | Power grid automation system and equipment running state risk assessment algorithm | |
CN115576284A (en) | Clothing workshop intelligent management method and system | |
CN106845826B (en) | PCA-Cpk-based cold continuous rolling production line service quality state evaluation method | |
CN117057644A (en) | Equipment production quality detection method and system based on characteristic matching | |
CN115698882A (en) | Abnormal modulation cause identification device, abnormal modulation cause identification method, and abnormal modulation cause identification program | |
CN111612149A (en) | Main network line state detection method, system and medium based on decision tree | |
CN113420061A (en) | Steady state working condition analysis method, optimization debugging method and system of oil refining and chemical production device | |
CN111931334A (en) | Method and system for evaluating operation reliability of cable equipment | |
CN115698881A (en) | Abnormal modulation cause identification device, abnormal modulation cause identification method, and abnormal modulation cause identification program | |
CN114548494B (en) | Visual cost data prediction intelligent analysis system | |
CN107103425B (en) | Intelligent energy evaluation system for power generation equipment running state computer | |
EP4206838A1 (en) | Forecasting and anomaly detection method for low density polyethylene autoclave reactor | |
CN117556366B (en) | Data abnormality detection system and method based on data screening | |
CN117469603B (en) | Multi-water-plant water supply system pressure optimal control method based on big data learning | |
CN110866558A (en) | Multi-source data fusion analysis-based rotating equipment state early warning method | |
CN111367255A (en) | Performance evaluation test system and method for multi-variable control system | |
CN116566839A (en) | Communication resource quality evaluation system for power enterprises |
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 |