CN113614758A - Equipment index goodness grade prediction model training method, monitoring system and method - Google Patents

Abstract

A training method of an equipment index goodness grade prediction model comprises the following steps: acquiring historical parameter values of a plurality of equipment indexes (S110); according to the incidence relation among the multiple equipment indexes, performing dimensionality reduction processing on the historical parameter values to obtain sample data about dimensionality reduction variables (S120); determining a positive sample set and a negative sample set according to the quality grade of the sample data of the dimension reduction variable (S130); and training a machine learning model through the positive sample set and the negative sample set to obtain a prediction model for predicting the goodness grade of the target equipment index (S140).

Description

Equipment index goodness grade prediction model training method, monitoring system and method Technical Field

The embodiment of the disclosure relates to the field of display panel manufacturing, and in particular relates to a training method of an equipment index goodness grade prediction model, a monitoring system and method of equipment indexes, and a computer readable medium.

Background

Obviously, the display panel needs to be processed by a plurality of processing devices in sequence during the production process. Meanwhile, the final display panel product inevitably has various defects with a certain probability, and the quality grade of the product is essentially related to process equipment, particularly caused by parameter values of equipment indexes.

Therefore, the method for determining the correlation between the equipment indexes of the process equipment and the quality grade of the product has important significance for poor positioning, production flow adjustment and the like.

Disclosure of Invention

The embodiment of the disclosure provides a training method of an equipment index goodness grade prediction model, a monitoring system and a monitoring method of equipment indexes, and a computer readable medium.

In an embodiment of the present disclosure, a training method of a device indicator goodness grade prediction model is provided, including:

acquiring historical parameter values of a plurality of equipment indexes;

according to the incidence relation among the multiple equipment indexes, performing dimensionality reduction processing on the historical parameter values to obtain sample data about dimensionality reduction variables;

determining a positive sample set and a negative sample set according to the quality grade of the sample data of the dimensionality reduction variable;

and training a machine learning model through the positive sample set and the negative sample set to obtain a prediction model for predicting the quality level of the target equipment index.

In an embodiment of the present disclosure, the performing, according to the correlation between the multiple device indicators, a dimension reduction process on the historical parameter value includes:

analyzing the association relation among the plurality of equipment indexes through a principal component analysis algorithm;

performing combined operation on the associated equipment indexes based on the association relation to obtain a dimension reduction variable;

and determining sample data of the dimensionality reduction variable according to the historical parameter value of the equipment index related to the combined operation.

In an embodiment of the disclosure, the training of the machine learning model by the positive sample set and the negative sample set includes:

respectively sampling the dimensionality reduction variables in the positive sample set and the dimensionality reduction variables in the negative sample set to obtain sampling variables;

calculating a feature vector of a fourier basis for each of said sampled variables to determine correlations between said sampled variables;

processing the characteristic vector of the Fourier basis based on a linear regression algorithm to obtain a screening variable;

and training a machine learning model based on the screening variables and the parameter values corresponding to the screening variables.

In an embodiment of the present disclosure, the sample data includes a training sample and a test sample;

the determining the positive sample set and the negative sample set according to the quality grade of the sample data of the dimension reduction variable includes:

determining a positive sample set and a negative sample set according to the quality grade of a training sample in the sample data of the dimensionality reduction variable;

the method further comprises the following steps: evaluating the predictive model through the test sample.

In one embodiment of the present disclosure, the prediction model is evaluated by one or more of the following evaluation indicators: accuracy, recall and area under receiver operating characteristic AUC.

In an embodiment of the present disclosure, the obtaining a historical parameter value of each of the equipment indexes includes:

loading source data regarding equipment metrics to a data warehouse for data mining analysis;

and storing the data after the mining analysis to a non-relational database for determining the historical parameter values.

In an embodiment of the present disclosure, the method further includes:

acquiring an index of equipment to be cleaned with parameter values being null values in the data object;

obtaining a mode value or an average value of attribute values of equipment indexes to be cleaned in other data objects;

and determining the mode value or the average value as the attribute value of the index of the equipment to be cleaned in the data object.

In an embodiment of the disclosure, the machine learning model is a bayesian algorithm model, a decision tree algorithm model, or a neural network model.

In an embodiment of the present disclosure, a system for monitoring an index of a device is provided, including: a distributed storage device, and one or more processors, wherein,

the distributed storage device is configured to respectively obtain first values of parameters to be measured related to M target device indicators, where M is an integer greater than 1;

the one or more processors are configured to monitor the goodness of the M target device metrics based on the predictive model.

In an embodiment of the disclosure, the one or more processors are further specifically configured to:

performing dimensionality reduction processing on the first parameter value to be measured according to the incidence relation among the M target equipment indexes to obtain a second parameter value to be measured related to N target dimensionality reduction variables, wherein N is a positive integer smaller than M;

and inputting the second parameter value to be measured into the prediction model so as to monitor the superiority of the indexes of the M target devices through the output value of the prediction model.

In an embodiment of the disclosure, the one or more processors are specifically configured to:

analyzing the incidence relation among the M target equipment indexes through a principal component analysis algorithm;

performing combined operation on the associated equipment indexes based on the association relation to obtain N target dimension reduction variables;

and determining a second parameter to be measured of the target dimension-reducing variable according to the first parameter to be measured of the target equipment index related to the combined operation.

In an embodiment of the disclosure, the one or more processors are specifically configured to:

inputting the second parameter values to be measured of the N target dimension-reducing variables into the prediction model, and respectively calculating the characteristic vectors of the Fourier bases corresponding to the N target dimension-reducing variables;

processing the feature vectors of the N Fourier bases based on a linear regression algorithm to obtain N 'target screening variables related to the goodness grade, wherein N' is smaller than N;

obtaining the prediction probability that the parameter values of the target screening variables respectively belong to each goodness grade based on the prediction model;

and determining the goodness grade corresponding to the maximum prediction probability value as the output value of the prediction model.

In an embodiment of the disclosure, the one or more processors are further configured to:

and if the output value of the prediction model is that the quality grade is lower than a preset grade value, performing dimension increasing processing on the target dimension reduction variable to obtain a fault equipment index set.

In an embodiment of the disclosure, the one or more processors are further configured to:

determining a good parameter value interval of a target equipment index to be monitored based on the prediction model, wherein the target equipment index to be monitored is any one of the M target equipment indexes;

and monitoring the current parameter value of the target equipment index to be monitored in real time according to the excellent parameter value interval.

In an embodiment of the present disclosure, the system further includes: a display device configured to:

and displaying a good parameter value interval related to the target equipment index to be monitored, and displaying the current parameter value of the target equipment index to be monitored, which is acquired in real time.

In an embodiment of the present disclosure, the system further includes: an alert device configured to:

and when the difference value between the current parameter value of the target equipment index to be monitored and the boundary value of the excellent parameter value interval is smaller than a preset value, giving an alarm.

In an embodiment of the present disclosure, a method for monitoring an equipment index is provided, including:

respectively acquiring first to-be-detected parameter values of M target equipment indexes, wherein M is an integer larger than 1;

the prediction model provided by the embodiment of the training method based on the equipment goodness grade prediction model monitors the goodness of the M target equipment indexes.

In an embodiment of the disclosure, the monitoring the superiority of the M target device indicators based on the prediction model includes:

performing dimensionality reduction processing on the first parameter value to be measured according to the incidence relation among the M target equipment indexes to obtain a second parameter value to be measured related to N target dimensionality reduction variables, wherein N is a positive integer smaller than M;

and inputting the second parameter value to be measured into the prediction model so as to monitor the superiority of the indexes of the M target devices through the output value of the prediction model.

In an embodiment of the present disclosure, the performing, according to the correlation between the M target device indicators, a dimension reduction process on the first parameter value to be measured includes:

analyzing the incidence relation among the M target equipment indexes through a principal component analysis algorithm;

performing combined operation on the associated equipment indexes based on the association relation to obtain N target dimension reduction variables;

and determining a second parameter to be measured of the target dimension-reducing variable according to the first parameter to be measured of the target equipment index related to the combined operation.

In an embodiment of the present disclosure, a computer readable medium is provided, on which a computer program is stored, wherein the program, when executed by a processor, implements a training method of a plant index goodness level prediction model as described above, and implements a monitoring method of a plant index as described above.

Drawings

The accompanying drawings are included to provide a further understanding of the embodiments of the disclosure and are incorporated in and constitute a part of this specification, illustrate embodiments of the disclosure and together with the description serve to explain the principles of the disclosure and not to limit the disclosure. The above and other features and advantages will become more apparent to those skilled in the art by describing in detail exemplary embodiments thereof with reference to the attached drawings, in which:

fig. 1 is a schematic flowchart of a training method of an equipment index goodness grade prediction model according to an embodiment of the present disclosure;

fig. 2 is a schematic flowchart of a method for obtaining a historical parameter value of an equipment indicator according to an embodiment of the present disclosure;

fig. 3 is a schematic flow chart of a dimension reduction processing method according to an embodiment of the disclosure;

FIG. 4 is a schematic flow chart diagram illustrating a model training method according to an embodiment of the present disclosure;

fig. 5 is a schematic structural diagram of a monitoring system for an equipment indicator according to an embodiment of the present disclosure;

fig. 6 is a schematic structural diagram of an analysis device in a monitoring system for device indicators according to an embodiment of the present disclosure;

fig. 7 is a schematic structural diagram of another monitoring system for equipment indicators according to an embodiment of the present disclosure;

fig. 8 is a schematic diagram of a monitoring manner of an equipment indicator according to an embodiment of the present disclosure;

FIG. 9 is a schematic diagram of a computer-readable medium provided by embodiments of the present disclosure;

fig. 10 is a schematic view of an electronic device provided in an embodiment of the present disclosure.

Detailed Description

In order to make those skilled in the art better understand the technical solutions of the embodiments of the present disclosure, the following describes in detail the system and method for product poor cause analysis and the computer readable medium provided by the embodiments of the present disclosure with reference to the drawings.

The disclosed embodiments will be described more fully hereinafter with reference to the accompanying drawings, but the illustrated embodiments may be embodied in different forms and should not be construed as limited to the embodiments set forth in the disclosure. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

Embodiments of the present disclosure may be described with reference to plan and/or cross-sectional views in light of idealized schematic illustrations of the present disclosure. Accordingly, the example illustrations can be modified in accordance with manufacturing techniques and/or tolerances.

Embodiments of the present disclosure and features of embodiments may be combined with each other without conflict.

The terminology used in the present disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. As used in this disclosure, the term "and/or" includes any and all combinations of one or more of the associated listed items. As used in this disclosure, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. The terms "comprises," "comprising," "made from … …," as used in this disclosure, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.

Unless otherwise defined, all terms (including technical and scientific terms) used in this disclosure have the same meaning as commonly understood by one of ordinary skill in the art. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and the present disclosure, and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

The disclosed embodiments are not limited to the embodiments shown in the drawings, but include modifications of configurations formed based on a manufacturing process. Thus, the regions illustrated in the figures have schematic properties, and the shapes of the regions shown in the figures illustrate specific shapes of regions of elements, but are not intended to be limiting.

Many products, such as display panels, are produced through production lines, each of which includes a plurality of process stations, each of which is used to perform certain processes (e.g., cleaning, deposition, exposure, etching, cartridge-to-cartridge, inspection, etc.) on the products, including semi-finished products. Meanwhile, each process station is provided with a plurality of process devices for carrying out the same treatment; of course, although the treatment is theoretically performed in the same manner, the actual treatment effect is not completely the same because of the difference in model, state, and the like between different process apparatuses.

The production process of each product needs to pass through a plurality of process stations, and the process stations passed by different products in the production process may be different; while products passing through the same process station may be processed by different process equipment therein. Therefore, in a production line, each process equipment participates in the production of a part of the products, but not all of the products, i.e., each process equipment participates in the production of only a part of the products.

In a first aspect, an embodiment of the present disclosure provides a training method for an equipment index goodness grade prediction model.

The prediction model of the equipment index goodness grade obtained by training through the training method provided by the embodiment of the disclosure can predict the parameter value of the equipment index, and the prediction parameter value is monitored in real time through the historical parameter trend, so as to predict the possibility of bad occurrence; the correlation between each process equipment in the production line and various defects of the product is determined, namely, the cause of the defects of the product is determined, so that the defects are positioned, the production flow is adjusted, and the like.

Further, the model for predicting the goodness grade of the equipment index obtained by training through the training method provided by the embodiment of the disclosure can also obtain an optimal trend graph of the parameter of each equipment index, so that problems can be found as early as possible during production of similar products, the equipment state is an optimal production state during production of the similar products, and the yield of production of the similar products is improved in batches.

"poor" refers to quality defects in the product that may result in a reduced quality or even scrapped product, or may result in a need for rework or repair of the product.

Among them, the defects can be classified into various types as required. For example, the product properties can be classified according to the direct effect of defects on the product properties, such as bright line defects, dark line defects, firefly defects (hot spots), and the like; or, the data lines can be classified according to specific causes of the defects, such as poor short circuit between the gate lines and the data lines, poor alignment, and the like; or, the method can also be classified according to the general causes of the defects, such as array process defect, color film process defect and the like; alternatively, the classification may be based on the severity of the defect, such as a defect that results in scrapping, a defect that results in reduced quality, etc.; alternatively, the type of defect may not be distinguished, i.e., a product is considered "bad" whenever any "defect" exists, and is considered "good" otherwise.

Of course, the above correlation is for "one" type of defect, i.e., the same process equipment has different correlations for different types of defects. The term "correlation between process equipment and a certain type of defect" means the degree of influence of the presence of the process equipment on the probability of occurrence of the defect in the product.

In some embodiments, the product is a display panel production line. The embodiment of the disclosure can be used for determining the relevance of each process device of a display panel production line and poor relevance in the production process of the display panel (such as a liquid crystal display panel, an organic light emitting diode display panel and the like). Of course, embodiments of the present disclosure may be used with other products as well.

Referring to fig. 1, the embodiment of the present disclosure provides a method for training a device indicator goodness grade prediction model, which includes steps S110 to S140.

In step S110, historical parameter values of a plurality of equipment indexes are acquired.

Wherein the device index is related to the quality of the product (e.g., OLED), and comprises: temperature, pressure, humidity, heating time, cooling time, etc. The device index may be an index of a plurality of devices or an index of one device. The historical parameter values of each of the equipment indexes include: the temperature value in the preset historical duration, the pressure value in the preset historical duration and the like.

In an exemplary embodiment, in the related art, a monitoring model is separately established for each equipment index to realize parameter monitoring of the equipment index, and further, parameter monitoring of each equipment index needs manual separate monitoring. However, more equipment indexes will result in more data of the monitoring model, and in order to save manpower and material resources, only some important parameters are managed, so that the monitoring range is smaller and the monitoring accuracy needs to be improved.

In order to overcome the problems, the technical scheme takes Factory Data Collection (FDC) system data as a big data basis for training the equipment index goodness grade prediction model. The FDC system is a system for monitoring and analyzing data (PROCESS data) of a production facility in real time, and can reduce a rejection rate and a production time and improve quality and yield of a product by monitoring and analyzing the data in real time, for example. Thus, the respective indexes of the equipment and the parameters thereof are collected. According to the scheme, the samples for training the classification model are determined based on the multiple equipment indexes and the historical parameter values of the equipment indexes, the monitoring range can be effectively expanded, and the monitoring accuracy is improved.

In an exemplary embodiment, the embodiment shown in fig. 2 provides a specific implementation manner of acquiring the historical parameter value of each equipment index in step S110, and includes:

step S210, loading source data about equipment indexes to a data warehouse (such as Hive) for data mining analysis; and, step S220, storing the data after the mining analysis to a non-relational database (e.g., Hbase) for determining the historical parameter values.

Specifically, the source data of the device indicator is derived from State Variable (SV) data in the FDC system. The SV data is used to represent real-time status parameters of the equipment index (e.g., the current temperature is 80 degrees celsius). In this embodiment, a BLOB (binary large object) field in the SV data may be parsed to generate the visualization data. The visual data refers to data which is transmitted to a user in a visual mode by means of a graphical means, so that information is transmitted and communicated more clearly and effectively.

In this embodiment, the SV data (which may include the above visualization data) with dispersion, disorder and non-uniform criteria may be integrated together through an ETL (Extraction-Transformation-Loading, i.e., a process of data Extraction (Extract), Transformation (Transform) and Loading (Load), which is an important link for constructing a data warehouse), and then imported into the data warehouse (e.g., HIVE), so as to provide an analysis basis for further data decision making. Therefore, massive structured full data can be stored, operated, mined and analyzed in a data warehouse.

Further, the data in the data warehouse HIVE is stored in a non-relational database (e.g., Hbase) through ETL, so as to store massive semi-structured data through Hbase and realize retrieval and real-time analysis of the data. Thereby, the historical parameter values of the equipment indexes can be respectively determined.

In an exemplary embodiment, to improve the prediction accuracy of the trained model, data cleaning is also performed when determining the sample set. Specifically, the method comprises the following steps:

acquiring an index of equipment to be cleaned with parameter values being null values in the data object; obtaining a mode value or an average value of attribute values of equipment indexes to be cleaned in other data objects; and determining the mode value or the average value as an attribute value of the index of the equipment to be cleaned in the data object.

Illustratively, the data stored in the above-mentioned non-relational database Hbase is washed. Taking the parameter value as a null value as an index of the equipment to be cleaned, and specifically, if the null value is numerical, filling the missing attribute value according to the average value of the values of the attribute in all other objects; if the null value is numerical, the missing value is supplemented by the value of the attribute with the maximum value taking times of all other objects according to a mode principle in statistics, so that the missing parameter value is supplemented by the value with the maximum probability possible, and the prediction accuracy of the trained model is improved.

Continuing to refer to fig. 1, in step S120, performing dimension reduction processing on the historical parameter values according to the association relationship among the multiple device indicators to obtain sample data about the dimension reduction variable.

In an exemplary embodiment, in the related art, a monitoring model is separately established for each equipment index to realize parameter monitoring of the equipment index, and further, the parameter monitoring of each equipment index needs to be separately monitored. Because a monitoring model is separately established for each equipment index to realize parameter monitoring of the equipment index, the equipment indexes lack correlation, and when a problem occurs in the target monitoring equipment, it is difficult to acquire parameter information of other equipment related to the target equipment parameter, resulting in low monitoring efficiency.

In order to overcome the problems, according to the technical scheme, dimension reduction processing is performed on relevant historical parameter values according to the incidence relation among equipment indexes, so that sample data about dimension reduction variables are obtained.

Specifically, fig. 3 is a flowchart illustrating a dimension reduction processing method according to an embodiment of the disclosure, which may be used as a specific implementation manner of step S120. Referring to fig. 3, the method of this embodiment includes:

step S310, analyzing the association relation among the multiple equipment indexes through a principal component analysis algorithm; step S320, performing combined operation on the associated equipment indexes based on the association relation to obtain a dimension reduction variable; and step S330, determining sample data of the dimensionality reduction variable according to the historical parameter value of the equipment index related to the combined operation.

In an exemplary embodiment, since there is a certain correlation between the device metrics, the device metrics having the correlation may be regarded as a set (i.e., "device metric set"). A set of device metrics may also be referred to as "Recipe". Specifically, a set of equipment index sets related to the completion of a certain function of a product in the manufacturing process of the product; for example, during OLED manufacturing, a set of device parameters (e.g., temperature, pressure, and electrical properties) are associated with the encapsulation stage. Also, the parameter values for a set of Recipe are constantly changing as technology improves. Then, taking the above mentioned Recipe related to the packaging stage as an example, before a certain packaging technology is improved, it may be referred to as a first version Recipe, and after the certain packaging technology is improved, it may be referred to as a second version Recipe. In this embodiment, reference basis is provided for optimizing the Recipe by comparing the difference of the parameter values in the Recipe between different versions.

Furthermore, quantitative statistical results are provided for each equipment index and different process steps in a set of Recipe, so that the equipment indexes (variables) with close relations are subjected to dimension reduction processing to be comprehensive indexes (namely, new variables with small quantity), and the obtained comprehensive indexes are irrelevant pairwise. Therefore, a small number of comprehensive indexes can be used for representing information contained in a large number of original equipment indexes.

The dimensionality reduction is to preserve the most important characteristics of the high-dimensional data and remove the unimportant characteristics, so that the purpose of improving the data processing speed is achieved, the dimensionality reduction is in a certain information loss range, a large amount of time and cost can be saved for people), the data is preprocessed, the high-dimensional data is mapped to a low-dimensional space, the variance of the data is maximized in the dimensionality, the characteristics of more original data points are preserved by using fewer data dimensionalities, and the inherent information of the data is maintained to the maximum extent.

Further, in step S130, a positive sample set and a negative sample set are determined according to the quality level of the sample data of the dimension-reduced variable.

In an exemplary embodiment, the quality level of the sample data is determined according to the quality level of the product, for example, the quality of the product is classified into a level a, B level, C level, and D level, and correspondingly, the sample data related to manufacturing the a level product may be classified into a level a, the sample data related to manufacturing the B level product may be classified into B level, the sample data related to manufacturing the C level product may be classified into C level, and the sample data related to manufacturing the D level product may be classified into D level. Further, the goodness levels of the training samples belonging to the class a and the class B in the sample data of the dimension-reduced variable may be determined as a positive sample set, and the goodness levels of the training samples belonging to the class C and the class D in the sample data of the dimension-reduced variable may be determined as a negative sample set.

Illustratively, the positive sample set and the negative sample set are determined with the final inspection data of the OLED product. Specifically, if the final inspection data of the OLED product is good, the equipment index parameter for producing the OLED product is used as a positive sample, so that a positive sample set is determined; and if the final inspection data of the OLED product is defective, the equipment index parameter for producing the OLED product is used as a negative sample, so that a negative sample set is determined. It should be noted that: in order to ensure the prediction accuracy, the positive and negative sample sets are determined for a single model of product. In addition, in order to avoid inaccurate prediction caused by uneven distribution of samples, the positive sample set at least includes 1 ten thousand pieces of positive sample information, and the negative sample set at least includes 1 ten thousand pieces of negative sample information for a single model of product.

Further, in step S140, a machine learning model is trained by the positive sample set and the negative sample set to obtain a classification model for predicting the quality level of the target device index.

In an exemplary embodiment, fig. 4 is a flowchart illustrating a method for training a model provided by the embodiment of the present disclosure, which may be used as a specific implementation manner of step S140. Referring to fig. 4, the method of this embodiment includes:

step S410, respectively sampling the dimensionality reduction variables in the positive sample set and the negative sample set to obtain sampling variables; step S420, calculating a feature vector of a Fourier base of each sampling variable to determine correlation among the sampling variables; step S430, processing the feature vector of the Fourier basis based on a linear regression algorithm to obtain a screening variable; and step S440, training a machine learning model based on the screening variables and the parameter values corresponding to the screening variables.

Illustratively, the set of positive samples is randomly divided into training positive samples and testing positive samples, and the set of negative samples is randomly divided into training negative samples and testing negative samples. The embodiment shown in this figure trains the machine learning model by training positive examples and training negative examples. The machine learning model is a Bayes algorithm model, a decision tree algorithm model or a neural network classification model.

Illustratively, in a parameter space corresponding to a training sample, a dimensionality reduction variable corresponding to an equipment index is sampled randomly to obtain a sampling variable. Then, a feature vector of a low-degree Fourier basis is calculated for the parameter values corresponding to each of the sampled variables to determine the correlation between the sampled variables. Further, the feature vectors of the fourier bases are processed based on a linear regression algorithm (e.g., a lasso algorithm) to obtain the screening variables. And finishing one round of screening, wherein screening variables are characteristics with large influence on the quality grade of the product. Based on at least one round of screening, the machine learning model is trained with the screened features and their corresponding parameters. Through the technical scheme shown in fig. 4, noise variables which do not affect the quality grade of the product greatly enough can be removed, and the model training speed can be improved.

In an exemplary embodiment, the technical solution further includes evaluating a goodness-grade prediction model (referred to as "model to be tested") by testing the positive sample and the negative sample, verifying a test result of the model to be tested using at least one test index, and performing goodness-grade prediction on the model meeting the test index.

Specifically, the performance of the model is evaluated through one or more of the following evaluation indexes: the prediction model after iterative optimization is evaluated by accuracy, recall rate, KS (Kolmogorov-Smirnov ) value and Area AUC (a model evaluation index specifically used for evaluating the prediction value of the model; short for Area Under ROC) of a Receiver operation Characteristic Curve (ROC).

In an exemplary embodiment, a method for testing a model to be tested specifically includes:

firstly, inputting the description characteristics of a test sample into a model to be tested, and obtaining the following output data of the model: true positive TP, true negative TN, false negative FN and false positive FP. The TP is the number of the positive classes in the test sample set judged by the to-be-tested model, the TN is the number of the negative classes in the test sample set judged by the to-be-tested model, the FN is the number of the negative classes in the test sample set judged by the to-be-tested model, and the FP is the number of the positive classes in the test sample set judged by the to-be-tested model. The positive class and the negative class refer to two classes manually labeled on the first part of samples, namely, if a certain sample is manually labeled to belong to a specific class, the sample belongs to the positive class, and the sample which does not belong to the specific class belongs to the negative class.

And secondly, calculating the test result of the model to be tested according to the true positive TP, the true negative TN, the false negative FN and the false positive FP.

In an exemplary embodiment, the test indicator can be an AUC or KS value. Specifically, the method comprises the following steps:

in an exemplary embodiment, the false positive rate FPR and the true positive rate TPR are determined using equations one and two,

FPR is equal to FP/(FP + TN) formula one

TPR is TP/(TP + FN) formula two

Further, the ROC curve is plotted with FPR as abscissa and TPR as ordinate. And the ROC curve is a characteristic curve of each obtained index, is used for displaying the relation among the indexes and further calculates the area AUC under the ROC curve. The ROC curve is a characteristic curve of each obtained index and is used for displaying the relation among the indexes, the AUC (area under the ROC curve) is larger, the prediction value of the model is higher, and the model to be tested can be tested through the AUC. And when the evaluation result is that the AUC value meets the preset threshold, the obtained model can be used for predicting the goodness grade.

KS (each time we choose a different threshold, we can get a set of FPR and TPR, i.e. a point on the ROC curve) curve is the difference between the two curves at each threshold. KS ═ max (TPR-FPR), which is the maximum value of the difference between TPR and FPR; the KS value can reflect the optimal distinguishing effect of the model, and the threshold value taken at the moment is generally used as the optimal threshold value for defining the quality of the user. The value range of the KS value is [0, 1], and the larger the KS value is, the higher the prediction accuracy of the model is. Illustratively, KS >0.2 indicates that the model has good prediction accuracy.

In an exemplary embodiment, the model test metrics may also be accuracy, recall. Specifically, the accuracy rate p and the recall rate r are respectively calculated according to a formula three and a formula four;

formula III

r is TP/(TP + FN) formula IV

If the test index corresponds to the setting conditions: if the accuracy test result is greater than p '(preset value), the accuracy setting condition is met, otherwise, the accuracy setting condition is not met, and if the recall rate test result is greater than r' (preset value), the recall rate setting condition is met, otherwise, the recall rate setting condition is not met.

In an exemplary embodiment, when the test result meets the setting condition corresponding to the test index, the model to be tested may be used as a prediction model for predicting the parameter value of the equipment index; and when the test result does not meet the set condition, continuing iterative optimization of the model to be tested until the test result of the model to be tested meets the set condition.

In a second aspect, an embodiment of the present disclosure provides a system for monitoring an equipment index.

Since the present invention relates to a plurality of plant devices of a plurality of plants, the data amount of raw data regarding the device index and the parameter value is large. For example, all plant equipment may produce hundreds of G of raw data per day and tens of G of data per hour. There are two main schemes for realizing storage and calculation of massive structured data: a grid computing scheme for RDBMS Relational Database Management (RDBMS); big data schema of Distributed File management System (DFS).

The grid calculation of the RDBMS divides a problem requiring a very large calculation power into many small parts, allocates the small parts to many computers for separate processing, and finally integrates the calculation results. For example, as a specific example, an Oracle RAC (real application cluster) is a core technology of grid computing supported by an Oracle database, in which all servers have direct access to all data in the database. However, the application system of the grid computing of the RDBMS cannot meet the user requirement when the data volume is large, for example, the efficiency of processing data is very low due to the input/output bottleneck of the hard disk after the data is increased to a large enough order because the expansion space of the hardware is limited.

Distributed file management based big data technology allows a large cluster to be constructed by using a plurality of cheap hardware devices to process massive data. For example, the Hive tool is a data warehouse tool based on Hadoop, and can be used for data Extraction-Transformation-Loading (ETL), and the Hive tool defines a simple SQL-like query language and allows a user-defined mapper and reducer of MapReduce to default complex analysis work that cannot be completed by the tool. The Hive tool has no special data storage format and does not establish indexes for data, and a user can freely organize a table in the Hive tool to process data in a non-relational database. Therefore, the parallel processing of the distributed file management can meet the storage and processing requirements of mass data, a user can query and process simple data through SQL, and a user-defined function can be adopted for complex processing. Therefore, in analyzing the mass data of the factory, the data of the factory database needs to be extracted into a distributed file system (such as the distributed storage device 501/701 described below), specifically:

and integrating the SV data with dispersion, disorder and non-uniform standards together through the ETL, and importing the SV data into a data warehouse HIVE to provide analysis basis for further data decision making. Therefore, massive structured full-scale data can be stored, operated, mined and analyzed in the data warehouse HIVE. Further, the data in the data warehouse HIVE is stored in a non-relational database Hbase through the ETL, so that massive semi-structured data can be stored through the Hbase, and data retrieval and real-time data analysis are achieved.

Therefore, on one hand, the original data cannot be damaged, and on the other hand, the data analysis efficiency is improved.

Referring to fig. 5, a monitoring system 500 for device indicators provided by embodiments of the present disclosure includes a distributed storage device 501 and one or more processors 502.

The distributed storage device 501 is configured to obtain first values of parameters to be measured related to M target device indicators, where M is an integer greater than 1; for storing production data generated by plant equipment. The one or more processors 502 are configured to monitor the goodness of the M target device metrics based on the classification model provided by the embodiments described above. An operation of determining a correlation is performed.

The distributed storage device 501 stores production data from plant equipment. The factory equipment refers to any equipment in each factory, and may include process equipment in each process site, management equipment for managing a production line in the factory, and the like; the production data refers to any information related to production, including which products each process equipment in each production line participates in, whether each product is finally bad or not, and what kind of bad exists.

Referring to FIG. 6, one or more processors (e.g., CPUs) 502 described above can be in analysis device 600. Illustratively, the analysis device 600 further includes a memory (e.g., a hard disk) 601 storing a desired program, and the processor 502 is connected to the memory 601 through an I/O interface 602 to enable information interaction, so that the processor 502 can perform desired operations according to the program stored in the memory 601 to achieve the operation of determining the correlation.

The distributed storage device 501 stores relatively complete data (e.g., a non-relational database), and the distributed storage device 501 includes a plurality of hardware memories, and different hardware memories are distributed at different physical locations (e.g., at different factories or different production lines), and communicate information with each other via a network, so that the data is in a distributed relationship, but logically forms a database based on big data technology.

Referring to fig. 7, a monitoring system 700 is shown, comprising: a distributed storage device 701, a source system 702, an analysis device 703 comprising one or more processors, and a display device 704.

The source system 702 includes raw data (e.g., temperature-related parameter values, pressure-related parameter values, electrical-related parameter values, etc.) for a number of different plant devices. Illustratively, the raw numbers may be processed by a PLC (Programmable Logic Controller). Further, the processed data is stored in a CIM PC (machine storing device related information), and then, the EIS (Equipment international System) classifies and transmits the required data from the CIM PC to further store in a relational database (e.g., Oracle, myql, etc.) of a plurality of production and Manufacturing systems, such as a YMS (Yield Management System) that stores Yield related data, an FDC System, an MES (Manufacturing Execution System), an MDW (Manufacturing data Warehouse), and the like. Further, the data are stored in a HADOOP big data platform in an ETL mode so as to perform correlation calculation on the HADOOP. For example, the data stored in the HADOOP big data platform may be further displayed in a corresponding report by a bo (business object) tool.

Further, data reading by the subsequent analysis equipment 703 is facilitated while load on factory equipment and production manufacturing systems is reduced. Data of the source System can be transmitted to the data lake 71 of the Distributed storage device (e.g. Hadoop Distributed File System, HDFSHadoop Distributed File System, HDFS)701 in real time by using OGG (short for Oracle Golden Gate, real-time data transmission tool of Oracle).

Data in distributed storage 701 may be stored in a data warehouse (e.g., Hive) or non-relational database (e.g., Hbase) format. Illustratively, data from the source system 702 is first stored in a data lake; after that, preprocessing such as data cleaning and data conversion can be continuously carried out in Hive according to application themes, scenes and the like of the data. Illustratively, the data warehouse 72 obtains different topics (such as production history topics, inspection data topics, point location measurement topics, and equipment data topics) through ETL processing and the like. Further, data with different scenarios (e.g., burst bad analysis scenario, correlation analysis scenario) are stored to the data mart 73. Illustratively, the data in the data marts 73 may be specifically stored in a non-relational database Hbase. The data marts 73 may be further connected to the analysis device 703, etc. through an API interface to implement data interaction with these devices.

For example, the data mart 73 transmits the collected parameter data corresponding to the equipment index to one or more processors in the analysis equipment 703 through an API interface, so as to perform data processing (see the embodiment shown in fig. 2 and 3 for specific implementation) to obtain a sample set; further, in one or more processors in the analysis device 703, model training is performed by training positive samples and training negative samples in the sample set (for a specific implementation, refer to the embodiment shown in fig. 4). In addition, model evaluation is performed by the positive test sample and the negative test sample in the sample set in one or more processors of the analysis device 703 (see the above examples of model test for specific embodiments).

In an exemplary embodiment, since the training samples of the prediction model are trained based on the dimensionality reduction variables obtained by performing the dimensionality reduction processing on the multiple device indexes, when the goodness level of the device index is predicted through the prediction model, the target device index (i.e., the device index to be tested) acquired in the distributed storage device 701 needs to be transmitted to the analysis device 703 to perform data processing (i.e., dimensionality reduction processing) on the target device index through one or more processors.

Specifically, in this embodiment, the parameter values corresponding to the M target device indicators are referred to as "first parameter values to be measured", and the parameter values corresponding to the N target dimension-reduced variables obtained after the dimension-reduction processing is performed on the M target device indicators are referred to as "second parameter values to be measured". The one or more processors in the analysis device 703 are further configured to: performing dimensionality reduction processing on the first parameter value to be measured according to the incidence relation among the M target equipment indexes to obtain a second parameter value to be measured related to N target dimensionality reduction variables, wherein N is smaller than a positive integer of M; and inputting the second parameter value to be measured into the prediction model so as to monitor the goodness of the M target equipment indexes through the output value of the prediction model.

Therefore, the technical scheme performs the dimension reduction processing according to the incidence relation between the target equipment indexes, and represents the information contained in the original equipment indexes with a large quantity through the comprehensive indexes with a small quantity.

The process of performing the dimension reduction processing on the M target device indicators by one or more processors in the analysis device 703 specifically includes: analyzing the incidence relation among the M target equipment indexes through a principal component analysis algorithm; performing combined operation on the associated equipment indexes based on the association relation to obtain N target dimension reduction variables; and determining a second parameter to be measured of the target dimension-reducing variable according to the first parameter to be measured of the target equipment index related to the combined operation.

It should be noted that, since the process of performing the above dimension reduction processing by one or more processors in the analysis device 703 is the same as that in the embodiment shown in fig. 3, no further description is given here.

Referring to fig. 7, the display device 704 may be used to display an "interactive interface," which may include a sub-interface for displaying the analysis result (e.g., the quality grade), a sub-interface for controlling the system for analyzing the product failure cause to perform the required work (e.g., task setting), and a sub-interface for controlling each process device (e.g., setting the process parameters thereof). That is, through the "interactive interface" of the display device, it is possible to achieve complete interaction (control and reception of results) of the user with the system for product failure cause analysis. Illustratively, the user may provide relevant parameters for the design of the algorithm in the analysis device 703 via the interactive interface, the user may also select a model for predicting the goodness level of the device metric (i.e., model selection) via the interactive interface, and so on.

In an exemplary embodiment, the goodness level prediction process for the target dimensionality reduction variable and the corresponding second parameter value to be measured by the trained prediction model is as follows, that is, one or more processors in the analysis device 703 are specifically configured to: inputting the second parameter values to be measured of the N target dimension reduction variables into the prediction model, and respectively calculating the characteristic vectors of the Fourier bases corresponding to the N target dimension reduction variables; processing the feature vectors of the N Fourier bases based on a linear regression algorithm to obtain N 'target screening variables related to the goodness grade, wherein N' is smaller than N; obtaining the prediction probability that the parameter values of the target screening variables respectively belong to each goodness grade based on the prediction model; and determining the goodness grade corresponding to the maximum prediction probability value as the output value of the prediction model.

In an exemplary embodiment, a specific implementation of determining the set of fault indicators is as follows: if the output value of the prediction model is that the goodness level is lower than the preset level value, the one or more processors in the analysis device 703 are further configured to: by performing dimension-up processing corresponding to the dimension-down processing on the target screening variable, an equipment index set (for example, it can be expressed as RecipeX) related to the target screening variable can be obtained. Further, a group of equipment indexes contained in the equipment index set RecipeX is determined as a fault equipment index set. Further, a correlation analysis (i.e., a correlation analysis topic shown in fig. 7) is performed on the set of equipment indexes included in the recipe of RecipeX to determine the cause of the undesirable occurrence. For example, if the quality level of the product is classified into a level a, a level B, a level C and a level D, the level B may be taken as the preset level value. Therefore, if the output values of the prediction model are in the C level and the D level, the equipment index set M related to the target screening variables is obtained to determine that the equipment index set M is a fault equipment index set. Of course, the preset grade value may also be a grade C, and then, when the output value of the prediction model is a grade D, the faulty device indicator set is determined. That is, the set of faulty equipment indicators is associated with the product quality levels divided in advance and the predicted values of the models.

In an exemplary embodiment, after the faulty equipment index set is determined, good parameter ranges corresponding to the equipment indexes in the index set are further determined. Specifically, since the current parameter value of at least one equipment index in the faulty equipment index set is out of the corresponding good parameter range, the current parameter value of the faulty equipment index can be adjusted according to the good parameter range, so that the faulty equipment index value can be timely restored to the corresponding good parameter range. Thereby reducing the reject ratio of the product.

It should be noted that, since the process of performing the above-mentioned data processing by one or more processors in the analysis device 703 is the same as that in the embodiment shown in fig. 4, it is not described herein again.

In an exemplary embodiment, the one or more processors in the analysis device 703 are further configured to: and determining a good parameter value interval of the target equipment index to be monitored based on the prediction model. Any one of the M target device indicators is used as the target device indicator to be monitored in this embodiment. Specifically, a good parameter value interval of the target equipment index to be monitored is determined according to the trained prediction model. Referring to fig. 8, the abscissa represents M target equipment indexes, and the ordinate represents a good parameter value interval for each target equipment index. Therefore, the current parameter value of the target equipment index to be monitored can be monitored in real time through the excellent parameter value interval.

Referring to FIG. 8, in an exemplary embodiment, a target device indicator Q for a target device to be monitored₁Obtaining Q in real time₁The parameter value of (2). When target equipment index Q to be monitored₁When the parameter value(s) of (1) is within the corresponding good parameter value interval, the description is given toThe above monitoring result of Q1 was good, which also indicates Q₁The current parameter value of (a) does not result in a poor product, in this embodiment Q is exemplified₁Is displayed as a dot. In addition, another target device index Q₂And the target device index Q to be monitored₁The same, for a target device index Q₂And will not be described in detail. In another exemplary embodiment, the metric Q is determined for a target device to be monitored_MObtaining Q in real time_MThe parameter value of (2). When target equipment index Q to be monitored_MWhen the parameter value(s) of (2) is outside the corresponding good parameter value range, the description will be given to the above-mentioned Q_MThe result of the monitoring of (A) is poor, and Q is also indicated_MMay result in a poor product, in this embodiment Q is illustratively set_MThe current parameter values are displayed as triangles to remind the staff to take improvement measures in time.

Therefore, according to the technical scheme, the influence of the equipment parameters on the product can be confirmed in a targeted manner through the history of the abnormal parameters in the whole life cycle process provided by big data prediction. And real-time monitoring is realized through the trend line of the variable parameters, so that the early warning of bad parameter values can be determined in time, and the reject ratio of products is reduced.

Referring to fig. 7, the display device 704 has a display function for displaying the analysis result (e.g., correlation) calculated by the analysis device.

In some embodiments, the display device further has a display function for displaying the analysis result (such as the goodness grade) calculated by the analysis device. The display device may include one or more displays including one or more terminals having a display function, so that the analysis device may transmit the correlation obtained by the analysis to the display device, and the display device may display the correlation.

In this embodiment, the display interface of the display device 704 is as shown in fig. 8: specifically, a good parameter value interval related to the target equipment index to be monitored is displayed, wherein the good parameter value interval refers to a parameter value range in which the parameter value of the target equipment index to be monitored does not cause generation of a bad product; and the display interface further displays the current parameter value of the target device index to be monitored, which is acquired in real time, wherein the current parameter value is displayed as a dot (which may be changed according to actual needs, for example, as a red dot) when the current parameter value is within the corresponding excellent parameter value interval, and is displayed as a triangular dot (which may be changed according to actual needs, for example, as a green dot) when the current parameter value is outside the corresponding excellent parameter value interval. Therefore, the good parameter value interval of any target equipment index and the image of the current parameter value are displayed as the staff, so that the staff can conveniently monitor the first-dimension variable and timely take corresponding measures for bad early warning.

In an exemplary embodiment, the monitoring system 700 further includes: and (5) an alarm device. In particular, the alarm device is configured to: and when the difference value between the current parameter value of the target equipment index to be monitored and the boundary value of the excellent parameter value interval is smaller than a preset value, giving out an alarm (such as a buzzer and a voice alarm lamp). The method and the device can remind workers of bad early warning, and then the workers can take corresponding measures timely for the bad early warning.

In the embodiment provided by the disclosure, the FDC system data is firstly used as a big data base for product quality grade prediction analysis, and further, the distributed storage device can efficiently collect and primarily process the raw data of a plurality of plant devices in a big data manner. One or more processors (which may be disposed in the analysis device) may then retrieve the required data from the distributed storage device. Based on the trained prediction model, good parameter value intervals of the M target equipment indexes can be determined, so that each target equipment index is monitored in real time, the probability of occurrence of failure and an alarm function are predicted, some important parameters which are not concerned before can be discovered to the FDC system through management of all parameters in failure prediction, the control range of the FDC system is enlarged, and therefore failure positioning, production flow adjustment and the like are accurately achieved. And the bad prediction can be automatically generated and provided for the FDC system, so that the FDC system is helped to reasonably confirm the range of related equipment indexes, the labor consumption is saved, and the problems can be found as early as possible. When the production of the similar products is ensured, the equipment state is the optimal production state, and the yield of the production of the similar products is improved in batch.

Referring to fig. 9, the present invention also provides a computer-readable medium 900 that can employ a portable compact disc read only memory (CD-ROM) and include program codes, and can be run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto, and in this document, a readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.

The program product described above may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a Random Access Memory (RAM), a Read-Only Memory (ROM), an erasable programmable Read-Only Memory (EPROM or flash Memory), an optical fiber, a portable compact disk Read-Only Memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

A computer readable signal medium may include a propagated data signal with readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A readable signal medium may also be any readable medium that is not a readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

Program code embodied on a readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Program code for carrying out operations for the present disclosure may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any kind of Network, including a Local Area Network (LAN) or Wide Area Network (WAN), or may be connected to an external computing device (e.g., through the internet using an internet service provider).

In addition, in an exemplary embodiment of the present disclosure, an electronic device capable of implementing the above method is also provided.

As will be appreciated by one skilled in the art, aspects of the present disclosure may be embodied as a system, method or program product. Accordingly, various aspects of the present disclosure may be embodied in the form of: an entirely hardware embodiment, an entirely software embodiment (including firmware, microcode, etc.) or an embodiment combining hardware and software aspects that may all generally be referred to herein as a "circuit," module "or" system.

An electronic device 1000 according to this embodiment of the disclosure is described below with reference to fig. 10. The electronic device 1000 shown in fig. 10 is only an example and should not bring any limitation to the functions and the scope of use of the embodiments of the present disclosure.

As shown in fig. 10, the electronic device 1000 is embodied in the form of a general purpose computing device. The components of the electronic device 1000 may include, but are not limited to: the at least one processing unit 1010, the at least one memory unit 1020, and a bus 1030 that couples various system components including the memory unit 1020 and the processing unit 1010.

The storage unit stores program codes, and the program codes can be executed by the processing unit 1010, so that the processing unit 1010 executes the method steps described in the embodiments of the present disclosure.

The storage unit 1020 may include readable media in the form of volatile storage units, such as: the Random Access Memory (RAM) 10201 and/or the cache Memory 10202 may further include a Read-Only Memory (ROM) 10203.

The memory unit 1020 may also include a program/utility 10204 having a set (at least one) of program modules 10205, such program modules 10205 including, but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which, or some combination thereof, may comprise an implementation of a network environment.

Bus 1030 may be any one or more of several types of bus structures including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, and a local bus using any of a variety of bus architectures.

The electronic device 1000 may also communicate with one or more external devices 1100 (e.g., keyboard, pointing device, bluetooth device, etc.), with one or more devices that enable a user to interact with the electronic device 1000, and/or with any devices (e.g., router, modem, etc.) that enable the electronic device 1000 to communicate with one or more other computing devices. Such communication may be through an Input/Output (I/O) interface 1050. Further, the I/O interface 1050 is connected to the display unit 1040 to transmit content to be displayed to the display unit 1040 through the I/O interface 1050 for viewing by a user.

Moreover, the electronic device 1000 may also communicate with one or more networks (e.g., a Local Area Network (LAN), a Wide Area Network (WAN), and/or a public Network such as the internet) via the Network adapter 1060. As shown, the network adapter 1060 communicates with the other modules of the electronic device 1000 over the bus 1030. It should be appreciated that although not shown, other hardware and/or software modules may be used in conjunction with the electronic device 1000, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, among others.

Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein may be implemented by software, or by software in combination with necessary hardware. Therefore, the technical solution according to the embodiments of the present disclosure may be embodied in the form of a software product, which may be stored in a non-volatile storage medium (which may be a CD-ROM, a usb disk, a removable hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which may be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the embodiments of the present disclosure.

Furthermore, the above-described figures are merely schematic illustrations of the processes involved in methods according to exemplary embodiments of the disclosure, and are not intended to be limiting. It will be readily understood that the processes shown in the above figures are not intended to indicate or limit the chronological order of the processes. In addition, it is also readily understood that these processes may be performed synchronously or asynchronously, e.g., in multiple modules.

Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any variations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.

Claims (20)

1. A training method of a device index goodness grade prediction model is provided, wherein the method comprises the following steps:

   acquiring historical parameter values of a plurality of equipment indexes;

   according to the incidence relation among the multiple equipment indexes, performing dimensionality reduction on the historical parameter values to obtain sample data about dimensionality reduction variables;

   determining a positive sample set and a negative sample set according to the quality grade of the sample data of the dimension reduction variable;

   and training a machine learning model through the positive sample set and the negative sample set to obtain a prediction model for predicting the goodness grade of the target equipment index.
2. The method of claim 1, wherein the performing the dimensionality reduction on the historical parameter values according to the correlation among the plurality of equipment indexes comprises:

   analyzing the incidence relation among the plurality of equipment indexes through a principal component analysis algorithm;

   performing combined operation on the associated equipment indexes based on the association relation to obtain a dimension reduction variable;

   and determining sample data of the dimension reduction variable according to the historical parameter value of the equipment index related to the combined operation.
3. The method of claim 1, wherein the training of a machine learning model by the set of positive samples and the set of negative samples comprises:

   respectively sampling the dimensionality reduction variables in the positive sample set and the dimensionality reduction variables in the negative sample set to obtain sampling variables;

   computing a feature vector of a fourier basis for each of the sampled variables to determine correlations between the sampled variables;

   processing the characteristic vector of the Fourier basis based on a linear regression algorithm to obtain a screening variable;

   and training a machine learning model based on the screening variables and the parameter values corresponding to the screening variables.
4. The method of any of claims 1 to 3, wherein the sample data comprises training samples and test samples;

   determining a positive sample set and a negative sample set according to the quality grade of the sample data of the dimension reduction variable, wherein the determining comprises the following steps:

   determining a positive sample set and a negative sample set according to the quality grade of a training sample in the sample data of the dimensionality reduction variable;

   the method further comprises the following steps: evaluating the predictive model through the test sample.
5. The method of claim 4, wherein the predictive model is evaluated by one or more of the following evaluation metrics: accuracy, recall and area under receiver operating characteristic AUC.
6. The method of any one of claims 1 to 3, wherein the obtaining of the historical parameter value for each of the equipment metrics comprises:

   loading source data regarding equipment metrics to a data warehouse for data mining analysis;

   storing data after the mining analysis to a non-relational database for determining the historical parameter values.
7. The method of claim 6, wherein the method further comprises:

   acquiring an index of equipment to be cleaned with parameter values being null values in the data object;

   obtaining a mode value or an average value of attribute values of equipment indexes to be cleaned in other data objects;

   and determining the mode value or the average value as an attribute value of the index of the equipment to be cleaned in the data object.
8. The method of any one of claims 1 to 3, wherein the machine learning model is a Bayesian algorithm model, a decision tree algorithm model, or a neural network model.
9. A system for monitoring a plant indicator, wherein the system comprises: a distributed storage device, and one or more processors, wherein,

   the distributed storage equipment is configured to respectively obtain first to-be-measured parameter values of M target equipment indexes, wherein M is an integer larger than 1;

   the one or more processors configured to monitor the goodness of the M target device metrics based on the predictive model of any of claims 1-8.
10. The system of claim 9, wherein the one or more processors are further specifically configured to:

    performing dimensionality reduction processing on the first parameter value to be measured according to the incidence relation among the M target equipment indexes to obtain a second parameter value to be measured related to N target dimensionality reduction variables, wherein N is smaller than a positive integer of M;

    and inputting the second parameter value to be measured into the prediction model so as to monitor the goodness of the M target equipment indexes through the output value of the prediction model.
11. The system of claim 10, wherein the one or more processors are specifically configured to:

    analyzing the incidence relation among the M target equipment indexes through a principal component analysis algorithm;

    performing combined operation on the associated equipment indexes based on the association relation to obtain N target dimension reduction variables;

    and determining a second parameter to be measured of the target dimension-reducing variable according to the first parameter to be measured of the target equipment index related to the combined operation.
12. The system of claim 11, wherein the one or more processors are specifically configured to:

    inputting the second parameter values to be measured of the N target dimension reduction variables into the prediction model, and respectively calculating the characteristic vectors of the Fourier bases corresponding to the N target dimension reduction variables;

    processing the feature vectors of the N Fourier bases based on a linear regression algorithm to obtain N 'target screening variables related to the goodness grade, wherein N' is smaller than N;

    obtaining the prediction probability that the parameter values of the target screening variables respectively belong to each goodness grade based on the prediction model;

    and determining the goodness grade corresponding to the maximum prediction probability value as the output value of the prediction model.
13. The system of claim 12, wherein the one or more processors are further configured to:

    and if the output value of the prediction model is that the quality grade is lower than a preset grade value, performing dimension increasing processing on the target dimension reduction variable to obtain a fault equipment index set.
14. The system of claim 9, wherein the one or more processors are further configured to:

    determining a good parameter value interval of a target equipment index to be monitored based on the prediction model, wherein the target equipment index to be monitored is any one of the M target equipment indexes;

    and monitoring the current parameter value of the target equipment index to be monitored in real time according to the excellent parameter value interval.
15. The system of claim 14, wherein the system further comprises: a display device configured to:

    displaying a good parameter value interval related to the target equipment index to be monitored, and displaying the current parameter value of the target equipment index to be monitored, which is acquired in real time.
16. The system of claim 14, wherein the system further comprises: an alert device configured to:

    and when the difference value between the current parameter value of the target equipment index to be monitored and the boundary value of the excellent parameter value interval is smaller than a preset value, giving an alarm.
17. A method for monitoring equipment indexes, wherein the method comprises the following steps:

    respectively acquiring first to-be-detected parameter values of M target equipment indexes, wherein M is an integer larger than 1;

    monitoring the goodness of the M target device metrics based on the predictive model of any one of claims 1-8.
18. The method of claim 17, wherein said monitoring goodness of the M target device metrics based on the predictive model comprises:

    performing dimensionality reduction processing on the first parameter value to be measured according to the incidence relation among the M target equipment indexes to obtain a second parameter value to be measured related to N target dimensionality reduction variables, wherein N is smaller than a positive integer of M;

    and inputting the second parameter value to be measured into the prediction model so as to monitor the goodness of the M target equipment indexes through the output value of the prediction model.
19. The method of claim 18, wherein the performing dimension reduction on the first value of the parameter to be measured according to the correlation between the M target device indicators includes:

    analyzing the incidence relation among the M target equipment indexes through a principal component analysis algorithm;

    performing combined operation on the associated equipment indexes based on the association relation to obtain N target dimension reduction variables;

    and determining a second parameter to be measured of the target dimension-reducing variable according to the first parameter to be measured of the target equipment index related to the combined operation.
20. A computer-readable medium, on which a computer program is stored, wherein the program, when executed by a processor, implements a method of training a plant index goodness level prediction model as claimed in any one of claims 1 to 8, or implements a method of monitoring a plant index as claimed in any one of claims 18 to 20.

Images