CN115619280A - Process quality prediction method based on process standard map and CNN-GRU network model - Google Patents

Process quality prediction method based on process standard map and CNN-GRU network model Download PDF

Info

Publication number
CN115619280A
CN115619280A CN202211369455.1A CN202211369455A CN115619280A CN 115619280 A CN115619280 A CN 115619280A CN 202211369455 A CN202211369455 A CN 202211369455A CN 115619280 A CN115619280 A CN 115619280A
Authority
CN
China
Prior art keywords
production
quality
cnn
quality prediction
method based
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
Application number
CN202211369455.1A
Other languages
Chinese (zh)
Inventor
蒲昊苒
阴艳超
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Kunming University of Science and Technology
Original Assignee
Kunming University of Science and Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Kunming University of Science and Technology filed Critical Kunming University of Science and Technology
Priority to CN202211369455.1A priority Critical patent/CN115619280A/en
Publication of CN115619280A publication Critical patent/CN115619280A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06395—Quality analysis or management
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
    • G06F16/367—Ontology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/04—Manufacturing
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Theoretical Computer Science (AREA)
  • Development Economics (AREA)
  • Physics & Mathematics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Physics & Mathematics (AREA)
  • Educational Administration (AREA)
  • Marketing (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Game Theory and Decision Science (AREA)
  • Quality & Reliability (AREA)
  • Operations Research (AREA)
  • Manufacturing & Machinery (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Animal Behavior & Ethology (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a process quality prediction method based on a process standard map and a CNN-GRU network model, and belongs to the technical field of process production. Vectorizing the triples in the process knowledge graph of the process production process; carrying out correlation analysis on the process parameters and the quality indexes of the process silk production line, and screening out the process parameters most relevant to the quality indexes; then carrying out tensor fusion on the screened process parameters and process index known word vectors in a known graph spectrum, and constructing a feature matrix by the formed composite production process features according to a time sliding window to serve as input variables of a prediction model; establishing a process quality prediction model by adopting a CNN-GRU neural network, and inputting the characteristics into the prediction model by combining a time sequence attention mechanism to obtain a predicted value of a quality index; and comparing and analyzing the prediction result and the true value of the process quality prediction model, and verifying the effectiveness of the method. The invention has important significance in improving the product quality and the manufacturing level in the production process.

Description

Process quality prediction method based on process standard map and CNN-GRU network model
Technical Field
The invention belongs to the technical field of process production, and particularly relates to a process quality prediction method based on a process standard map and a CNN-GRU network model.
Background
In the process manufacturing industry, the stability of each quality indicator during processing has a significant impact on the quality of the manufacturing product. Abnormal process parameters in the production process have great influence and fluctuation on the product quality and the production quantity of the production line. In order to deal with the quality index of the abnormity generated in the production process, the quality index is adjusted according to the requirements of the subsequent procedures by mainly utilizing process parameters, and the process quality index is accurately regulated and controlled to meet the standard range of the process requirements.
In the process of flow production and manufacturing, the fluctuation of the moisture content of the discharged material is large due to inaccurate setting of various influencing factors, and uncertainty, high coupling, nonlinearity and hysteresis exist among the processes. The model research for quality prediction at home and abroad is divided into two types: one is a method for analyzing time series by using historical data generated in the process of process production, the basic idea is that the historical data generated in the process production has time sequence, and the process quality at the next moment is predicted by using the data generated in a certain period of time, so that the method has the advantages of considering the time sequence relation of the data and having limited capability of predicting nonlinear relation data; the other is a machine learning analysis method, which utilizes the time sequence and nonlinear relation of data at the same time and is gradually applied to the quality prediction field. But text data such as some unstructured process experience knowledge, process standards, process grades and the like generated in the process production process are not considered, the technical standards and the process specifications are used as constraint conditions, and the method disclosed by the invention integrates multi-source heterogeneous data to predict the process quality, so that the defects of single data structure and knowledge loss are overcome.
With the progress of the times, workshop production tends to be highly intelligent, so that the production complexity in the process of flow manufacturing is increased sharply, the difficulty of improving production and processing is increased, the stability of quality indexes is difficult to ensure, the difficulty of predicting the quality indexes of a workshop is increased sharply, and problems exist in the process of flow manufacturing and production, for example, field operators are difficult to accurately regulate and control the water content of discharged materials, so that the waste of raw materials in the process of flow manufacturing is caused, the economic benefits of enterprises are reduced, and the like.
Disclosure of Invention
The invention aims to provide a process quality prediction method based on a process standard map and a CNN-GRU network model, which can accurately predict process quality, reduce the economic cost of enterprises, reduce the loss of raw materials and improve the manufacturing level of products.
The technical purpose of the invention is realized by the following technical scheme:
a process quality prediction method based on a process standard map and a CNN-GRU network model comprises the following steps:
step 1: vectorizing triples in the process production knowledge graph through word2vec words, and expressing process knowledge in the process knowledge graph through low-dimensional word vectors;
step 2: acquiring process data, namely acquiring and recording historical data generated in the process production process by using a sensor to obtain an initial variable;
and 3, step 3: screening process data, namely screening initial variables in the process production process by analyzing the correlation of historical data of each process in the process production, and obtaining production process parameters with the highest correlation with the quality index of the production process after removing miscellaneous variables with lower process correlation;
and 4, step 4: fusing a low-latitude word vector of process index knowledge in the knowledge map and a production process parameter with highest correlation of the production process quality index through tensor products to form input characteristics of a process quality prediction model;
and 5: establishing a process quality prediction model by adopting a CNN-GRU neural network, and inputting the input characteristics obtained in the step (4) into the process quality prediction model by combining a time sequence attention mechanism to perform numerical prediction of a process quality index;
and 6: and comparing and analyzing the prediction result of the process quality prediction model and the actual value of the process quality, and verifying the effectiveness and accuracy of the prediction result of the process quality prediction model.
Specifically, in step 1, word vectorization is performed on all phrases and characters forming the phrases in the knowledge-graph triple through word2vec, and then a new word vector containing word and character features is combined through corresponding word vectors and word vectors. The formula is as follows:
Figure BDA0003924393920000021
in the formula: v' words The new word vector is obtained after the character vector and the word vector are fused;
Figure BDA0003924393920000022
is the ith character vector; m represents a word with m characters; v words Is a word vector.
Specifically, the method comprises the following steps: in the step 2, the initial variables include historical data of the process processing process generated by 24 sets of process parameters (flow of the loosening and dampening process, hot air temperature of the loosening and dampening process, flow of water added for loosening and dampening and the like) and 1 set of quality indexes (discharged water content) in the process production.
Specifically, the method comprises the following steps: in the step 3, correlation analysis is performed on historical data of process parameters and quality indexes acquired by process production, so that 6 most relevant characteristics are obtained after process correlation degree and miscellaneous process parameter variables are removed, and a calculation formula is as follows:
Figure BDA0003924393920000031
in the formula: x is a production process parameter, and Y is a production process quality index; cov (X, Y) is the covariance between the two process variables; sigma X 、σ Y Is X, Y.
Specifically, the method comprises the following steps: 6 characteristic variables were obtained by correlation analysis: the method comprises the steps of loosening and moisture regaining process hot air temperature, loosening and moisture regaining water adding flow, primary feeding material liquid accumulation amount, primary feeding material accumulation amount, secondary feeding material accumulation amount and airflow drying process flow.
Specifically, the method comprises the following steps: in the step 4, the 6 characteristic variables and the word vectors corresponding to the process knowledge are fused through tensor to form new composite process production characteristics. The feature of the method is based on a calculation formula of tensor fusion:
Figure BDA0003924393920000032
in the formula: x is the number of v Producing a token, V ', for a composite process that fuses term vectors and token variables' words Is a word vector; x is a production process parameter;
Figure BDA0003924393920000033
is the tensor product.
Specifically, the method comprises the following steps: in the step 5, the CNN-GRU neural network includes an input layer, a flat layer, a convolutional layer, a pooling layer, a flat layer, an attention layer, four GRU layers, a full-link layer, and an output layer.
Specifically, the method comprises the following steps: in the step 5, a time sequence attention mechanism is combined, and weights are given to the input feature vectors at different process production moments, wherein the weights are given by the following formula:
Figure BDA0003924393920000034
in the formula: x' t Representing the importance value of the composite characteristic of the process parameter;
Figure BDA0003924393920000035
representing a composite feature of the input;
Figure BDA0003924393920000036
and representing the parameter weight corresponding to the input feature.
Specifically, the method comprises the following steps: in step 6, when the prediction result of the process quality prediction model is compared and analyzed with the actual process quality value, the Mean Absolute Error (MAE) and the Root Mean Square Error (RMSE) are used as the measurement standards of the process quality prediction model.
Specifically, the method comprises the following steps: the calculation method of the process quality prediction model measurement method comprises the following steps:
mean Absolute Error (MAE):
Figure BDA0003924393920000037
root Mean Square Error (RMSE):
Figure BDA0003924393920000041
in the formula: k is the number of process quality predicted values at all times; y is i And
Figure BDA0003924393920000044
respectively, the real value and the predicted value of the process quality at the time i.
In conclusion, the beneficial effects of the invention are as follows:
(1) And multi-source heterogeneous data generated in the process production process is fully utilized, and the integrity of process knowledge is completed.
(2) The method has the advantages that a production process quality prediction model is established, the problem of low prediction precision in the process production process is solved in a targeted manner, the stability requirement in the production process is met through accurate prediction of the process quality, and the quality of products is improved.
(3) Provides a method and a way for realizing accurate prediction and optimized regulation and control in the process production process.
Drawings
FIG. 1 is a schematic structural diagram of the present invention.
Fig. 2 is a partial word vector.
Fig. 3 is a schematic diagram of the internal structure of the GRU neural network of the present invention.
FIG. 4 is a schematic diagram of an internal structure of the timing attention mechanism of the present invention.
FIG. 5 is a comparison of predicted values for different models of the present invention.
Detailed Description
The invention will be further described with reference to the following figures and examples, without however restricting the scope of the invention thereto.
Example 1: referring to fig. 1 to 5, the method for predicting process quality based on a process standard map and a CNN-GRU network model disclosed by the invention comprises the following steps:
step 1: and performing word vectorization on all phrases and characters forming the phrases in the knowledge graph triple through word2vec, and then combining a new word vector containing the characteristics of the words and the characters through corresponding word vectors and word vectors. The formula is as follows:
Figure BDA0003924393920000042
in the formula: v' words The new word vector is obtained after the character vector and the word vector are fused;
Figure BDA0003924393920000043
is the ith character vector; m represents a word with m characters; v words Is a word vector.
Step 2: acquiring process Data, establishing a corresponding Data interface by a sensor, an SCADA (Supervisory Control and Data Acquisition) System and an MES (Manufacturing Execution System), recording and storing factor historical variable values related to process quality generated in a process production process, and acquiring initial variables; in the step 2, the initial variables include historical data of the process processing process generated by 24 sets of process parameters (flow of the loosening and dampening process, hot air temperature of the loosening and dampening process, flow of water added for loosening and dampening and the like) and 1 set of quality indexes (discharged water content) in the process production.
And 3, step 3: screening process data, screening initial variables through Pearson correlation analysis, selecting process parameters in process production variables with large Pearson correlation coefficients, and obtaining 6 most relevant characteristic variables after removing redundant process parameters: the method comprises the steps of loosening and moisture regaining process hot air temperature, loosening and moisture regaining water adding flow, primary feeding material liquid accumulation amount, primary feeding material accumulation amount, secondary feeding material accumulation amount and airflow drying process flow. The Pearson correlation coefficient calculation formula is as follows:
Figure BDA0003924393920000051
in the formula: x is a production process parameter, and Y is a production process quality index; u. of X u Y Respectively X and Y average values; cov (X, Y) is the covariance between the two; sigma X 、σ Y X, Y standard deviations, respectively.
And 4, step 4: and fusing 6 characteristic variables of the production process and word vectors corresponding to the process knowledge through tensor to form new composite process production characteristics. The method is based on a tensor fusion calculation formula:
Figure BDA0003924393920000052
in the formula: x is the number of v Producing a token, V ', for a composite process that fuses term vectors and token variables' words Is a word vector; x is a production process parameter;
Figure BDA0003924393920000053
is a tensor product.
And 5: establishing a prediction model by adopting a CNN-GRU neural network, and inputting input variables of the model by combining a time sequence attention mechanism to obtain prediction data of the water content of the discharged material; the CNN-GRU neural network comprises an input layer, a flat layer, a convolution layer, a pooling layer, a flat layer, an attention layer, four GRU layers, a full-connection layer and an output layer.
Step 6: and when the prediction result of the process quality prediction model is compared and analyzed with the process quality true value, the average absolute error (MAE) and the Root Mean Square Error (RMSE) are used as the measurement standards of the process quality prediction model.
The calculation method of the process quality prediction model measurement method comprises the following steps:
mean Absolute Error (MAE):
Figure BDA0003924393920000061
root Mean Square Error (RMSE):
Figure BDA0003924393920000062
in the formula: k is the number of process quality predicted values at all times; y is i And
Figure BDA0003924393920000064
respectively, the real value and the predicted value of the process quality at the time i.
The specific implementation process of the embodiment is as follows:
data are collected by a sensor and an SCADA (supervisory control and data acquisition) device in a certain production process and transmitted to an MES (manufacturing execution system), wherein the collection time is 2 months in 2021 to 9 months in 2021, and ten thousand and eight thousand sets of data are collected.
When the collected historical data of the production process is preprocessed, the average value of the historical data collected by the process is used for filling missing or wrong process values during collection, noisy point data is replaced or eliminated, and at the moment, seventy-thousand five-hundred historical data are used as a data set. And calculating correlation coefficients between 24 characteristic data and the quality indexes of the production process by using Pearson correlation, then evaluating and screening, wherein the calculation formula of the Pearson correlation coefficient is as follows, and selecting 6 characteristic variables with larger Pearson correlation coefficients.
Figure BDA0003924393920000063
In the formula: x is a production process parameter, and Y is a production process quality index; u. of X u Y Respectively X and Y average values; cov (X, Y) is covariance between them; sigma X 、σ Y X, Y standard deviations, respectively.
Through calculation and analysis of the correlation, six characteristics with large correlation with the water content of the discharged tobacco shred of the quality index are respectively as follows: the method comprises the steps of loosening and moisture regaining process hot air temperature, loosening and moisture regaining water adding flow, primary feeding material liquid accumulation amount, primary feeding material accumulation amount, secondary feeding material accumulation amount and airflow drying process flow. Therefore, when performing feature fusion, the above 6 feature variables are selected and fused.
According to the filamentation process standard corresponding to 6 characteristic variables analyzed by the correlation calculation of the process production historical data in the knowledge graph, the standard is expressed by word2vec vectorization, the vectorization result is shown in figure 2, and the vectorization result is fused with 6 production process variables analyzed by Pearson correlation, and the fusion formula is as follows:
Figure BDA0003924393920000071
in the formula: x is a radical of a fluorine atom v Producing a token, V ', for a composite process that fuses term vectors and token variables' words Is a word vector; x is a characteristic variable;
Figure BDA0003924393920000072
is the tensor product.
After the fusion, the dimension of each composite process feature is different, and in order to make the dimension the same, normalization processing needs to be performed on each dimension of data, wherein a normalization formula is as follows:
Figure BDA0003924393920000073
in the formula:
Figure BDA0003924393920000074
is a normalized value, x i For each dimension of the process features, x min 、x max The minimum value and the maximum value in each dimension data are respectively.
The invention uses 18000 pieces of historical data generated in the production process of silk making to be fused with relevant standards in the knowledge map of the silk making process, and the first 80 percent of the data set after the characteristics are fused is selected as a training set, and the last 20 percent is selected as a testing set.
The word vectorization dimension of relevant knowledge in the knowledge graph is 10, the time sequence of the quality index is predicted, so that a sliding window of data is set, the window size is set to be 60, the input variable characteristic quantity is 6, each characteristic dimension is 22, and based on the setting, the data of the last six minutes is used as a prediction model to be input, and the numerical value of the moisture content of the spinning discharging material in the next minute is predicted. Thus obtaining the final predicted water content prediction values of 35000 cut tobacco making discharge materials.
And predicting the water content of the discharged material of the production process quality index by using the CNN-GRU process quality prediction model of the processed composite characteristic variable. The CNN-GRU process quality prediction model comprises an input layer; a flat layer; a layer of convolution layer, the number of neurons is 256, the size of convolution kernel is 3, and the activation function is Relu; a layer of pooling layer, the size of the filter is 3; a flat layer; an attention layer; four GRU layers, the number of neurons is 128, and an activation function is Relu; a full connection layer, the number of neurons is 32, and an activation function is Relu; an output layer, the number of neurons is 1, the activation function is Relu, and the output layer is used for outputting the final predicted value, wherein the structural diagrams of GRU and the time sequence attention mechanism are shown in FIG. 3 and FIG. 4, and the structural diagrams are shown in the figure
Figure BDA0003924393920000075
The weights before the normalization are represented by,
Figure BDA0003924393920000076
representing the weights after normalization, xt representing the input features at time t, ht-1 representing the state left at the last time, rt representing the reset gate, zt representing the update gate,
Figure BDA0003924393920000077
represents the candidate state at time t, and σ and tanh represent activation functions.
And (3) inputting the fused process data to predict the water content of the process quality cut tobacco by using a process quality prediction model established by a neural network.
The model of the invention is used for predicting the predicted value of the process quality discharge water content and the corresponding true value to be compared and analyzed, as shown in figure 5; in order to judge the accuracy of the invention, the MAE and RMSE values of the process quality prediction model are calculated, the smaller the MAE and RMSE values are, the better and more effective the accuracy of the process prediction model is, and the smaller the error value is, as shown in Table 1, the process standard map and the CNN-GRU neural network process prediction model have very good prediction accuracy.
TABLE 1 evaluation index of different models
Figure BDA0003924393920000081
What has been described above are merely some of the embodiments of the present invention. It will be apparent to those skilled in the art that various changes and modifications can be made without departing from the inventive concept thereof, and these changes and modifications can be made without departing from the spirit and scope of the invention.

Claims (9)

1. A process quality prediction method based on a process standard map and a CNN-GRU network model is characterized by comprising the following steps:
step 1: vectorizing triples in the process production knowledge graph through word2vec words, and expressing process knowledge in the process knowledge graph through low-dimensional word vectors;
step 2: acquiring process data, namely acquiring and recording historical data generated in the process production process by using a sensor to obtain an initial variable;
and step 3: screening process data, namely screening initial variables in the process production process by analyzing the correlation of historical data of each process in the process production, and obtaining production process parameters with the highest correlation with quality indexes in the production process after removing redundant variables with lower process correlation;
and 4, step 4: fusing low-latitude word vectors of process index knowledge in the process knowledge in the knowledge map and production process parameters with highest correlation of production process quality indexes through tensor products to form input characteristics of a process quality prediction model;
and 5: establishing a process quality prediction model by adopting a CNN-GRU neural network, and inputting the input characteristics obtained in the step (4) into the process quality prediction model by combining a time sequence attention mechanism to carry out numerical prediction on process quality indexes;
step 6: and comparing and analyzing the prediction result of the process quality prediction model and the actual value of the process quality, and verifying the effectiveness and accuracy of the prediction result of the process quality prediction model.
2. The process quality prediction method based on the process standard map and the CNN-GRU network model as claimed in claim 1, wherein: in the step 1, word vectorization is performed on all phrases and characters forming the phrases in the knowledge graph triple through word2vec, and then a new word vector containing word and character features is formed through corresponding word vectors and character vectors, wherein the formula is as follows:
Figure FDA0003924393910000011
in the formula: v w ' ords The new word vector is obtained after the character vector and the word vector are fused;
Figure FDA0003924393910000012
is the ith character vector; m represents a word with m characters; v words Is a word vector.
3. The process quality prediction method based on the process standard map and the CNN-GRU network model as claimed in claim 1, wherein: in step 2, the initial variables include the historical process data generated by 24 sets of process parameters and 1 set of quality indexes in the process production.
4. The process quality prediction method based on the process standard graph and the CNN-GRU network model according to claim 3, wherein: in the step 3, correlation analysis is performed on historical data of process parameters and quality indexes acquired by process production, so that 6 most relevant characteristics are obtained after low process correlation degree and miscellaneous process parameter variables are removed, and a calculation formula is as follows:
Figure FDA0003924393910000021
in the formula: x is a production process parameter, and Y is a production process quality index; cov (X, Y) is the covariance between the two process variables; sigma X 、σ Y Is X, Y standard deviation.
5. The process quality prediction method based on the process standard map and the CNN-GRU network model as claimed in claim 4, wherein: in the step 4, the 6 characteristic variables and the word vectors corresponding to the process knowledge are fused through tensor to form new composite process production characteristics, and the calculation formula of the characteristics based on tensor fusion is as follows:
Figure FDA0003924393910000022
in the formula: x is a radical of a fluorine atom v Producing features for a composite process incorporating term vectors and feature variables, V w ' ords Is a word vector; x is a production process parameter;
Figure FDA0003924393910000023
is the tensor product.
6. The process quality prediction method based on the process standard graph and the CNN-GRU network model according to claim 1, wherein: in the step 5, the CNN-GRU neural network includes an input layer, a flat layer, a convolutional layer, a pooling layer, a flat layer, an attention layer, four GRU layers, a full-link layer, and an output layer.
7. The process quality prediction method based on the process standard graph and the CNN-GRU network model according to claim 1, wherein: in the step 5, a time sequence attention mechanism is combined, and weights are given to the input feature vectors at different process production moments, wherein the weights are given by the following formula:
Figure FDA0003924393910000024
in the formula: x is the number of t ' represents a composite characteristic significance value of a process parameter;
Figure FDA0003924393910000025
representing a composite feature of the input;
Figure FDA0003924393910000026
and representing the parameter weight corresponding to the input feature.
8. The process quality prediction method based on the process standard map and the CNN-GRU network model as claimed in claim 1, wherein: in step 6, when the prediction result of the process quality prediction model is compared and analyzed with the actual process quality value, the average absolute error MAE and the root mean square error RMSE are used as the measurement standards of the process quality prediction model.
9. The process quality prediction method based on the process standard graph and the CNN-GRU network model according to claim 8, wherein: the calculation method of the process quality prediction model measurement method comprises the following steps:
mean absolute error MAE:
Figure FDA0003924393910000031
root mean square error RMSE:
Figure FDA0003924393910000032
in the formula: k is the number of process quality predicted values at all times; y is i And
Figure FDA0003924393910000033
respectively, the real value and the predicted value of the process quality at the time i.
CN202211369455.1A 2022-11-03 2022-11-03 Process quality prediction method based on process standard map and CNN-GRU network model Pending CN115619280A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211369455.1A CN115619280A (en) 2022-11-03 2022-11-03 Process quality prediction method based on process standard map and CNN-GRU network model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211369455.1A CN115619280A (en) 2022-11-03 2022-11-03 Process quality prediction method based on process standard map and CNN-GRU network model

Publications (1)

Publication Number Publication Date
CN115619280A true CN115619280A (en) 2023-01-17

Family

ID=84876098

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211369455.1A Pending CN115619280A (en) 2022-11-03 2022-11-03 Process quality prediction method based on process standard map and CNN-GRU network model

Country Status (1)

Country Link
CN (1) CN115619280A (en)

Similar Documents

Publication Publication Date Title
CN111461555A (en) Production line quality monitoring method, device and system
CN116757534B (en) Intelligent refrigerator reliability analysis method based on neural training network
CN108875772B (en) Fault classification model and method based on stacked sparse Gaussian Bernoulli limited Boltzmann machine and reinforcement learning
CN110543616B (en) SMT solder paste printing volume prediction method based on industrial big data
CN108875771B (en) Fault classification model and method based on sparse Gaussian Bernoulli limited Boltzmann machine and recurrent neural network
CN107679715A (en) A kind of electric energy meter comprehensive error process merit rating method and evaluation system based on SPC
CN112486111B (en) Edible oil alkali refining process intelligent adjusting method based on data analysis
CN113743016B (en) Engine residual life prediction method based on self-encoder and echo state network
CN114282443B (en) Residual service life prediction method based on MLP-LSTM supervised joint model
CN110458616A (en) A kind of finance product recommended method based on GAMxNN model
CN115373370A (en) Method and system for monitoring running state of programmable controller
CN115438897A (en) Industrial process product quality prediction method based on BLSTM neural network
CN113627594B (en) One-dimensional time sequence data augmentation method based on WGAN
CN114692507A (en) Counting data soft measurement modeling method based on stacking Poisson self-encoder network
CN114548494A (en) Visual cost data prediction intelligent analysis system
CN114358116A (en) Oil-immersed transformer fault diagnosis method and system and readable storage medium
CN115619280A (en) Process quality prediction method based on process standard map and CNN-GRU network model
CN116933643A (en) Intelligent data monitoring method based on partial robust M regression and multiple interpolation
CN116662925A (en) Industrial process soft measurement method based on weighted sparse neural network
CN114741876B (en) Intelligent inspection method for tower crane
CN115169737A (en) Process quality prediction method based on CNN-LSTM hybrid neural network model
CN111062118B (en) Multilayer soft measurement modeling system and method based on neural network prediction layering
CN113849479A (en) Comprehensive energy supply station oil tank leakage detection method based on instant learning and self-adaptive threshold
CN114298413A (en) Hydroelectric generating set runout trend prediction method
CN113111588A (en) NO of gas turbineXEmission concentration prediction method and device

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