CN110569966A - Data processing method and device and electronic equipment - Google Patents
Data processing method and device and electronic equipment Download PDFInfo
- Publication number
- CN110569966A CN110569966A CN201910848176.5A CN201910848176A CN110569966A CN 110569966 A CN110569966 A CN 110569966A CN 201910848176 A CN201910848176 A CN 201910848176A CN 110569966 A CN110569966 A CN 110569966A
- Authority
- CN
- China
- Prior art keywords
- index
- data
- sample
- sample data
- training
- 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.)
- Granted
Links
- 238000003672 processing method Methods 0.000 title abstract description 8
- 238000012549 training Methods 0.000 claims abstract description 156
- 238000000034 method Methods 0.000 claims abstract description 38
- 238000012545 processing Methods 0.000 claims abstract description 26
- 230000007246 mechanism Effects 0.000 claims description 18
- 238000010801 machine learning Methods 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 description 14
- 238000010586 diagram Methods 0.000 description 10
- 238000004519 manufacturing process Methods 0.000 description 9
- 230000003247 decreasing effect Effects 0.000 description 4
- 238000005457 optimization Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000002360 preparation method Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000001172 regenerating effect Effects 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Strategic Management (AREA)
- Artificial Intelligence (AREA)
- Economics (AREA)
- Evolutionary Computation (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Development Economics (AREA)
- Educational Administration (AREA)
- Entrepreneurship & Innovation (AREA)
- Primary Health Care (AREA)
- Manufacturing & Machinery (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application discloses a data processing method, a data processing device and electronic equipment, wherein the method comprises the following steps: obtaining at least one training sample, wherein the training sample comprises index data of at least one index, the index data comprises sample data of the index of the training sample at least one moment, and the training sample has a preset yield label; denoising the sample data in the training sample based on the time attribute and the index attribute corresponding to the sample data; training the initial model based on the denoised training sample and the yield label to obtain a relation model, wherein the relation model represents the relation between the index and the yield. Therefore, the sample data is denoised by the time attribute and the index attribute in the sample data, so that the sample data used for training the relational model is more accurate, the relational model trained by the sample data is more accurate, and the aim of improving the accuracy of the relational model is fulfilled.
Description
Technical Field
The present application relates to the field of model training technologies, and in particular, to a data processing method and apparatus, and an electronic device.
Background
For the manufacturing industry, the process index of the equipment directly determines the yield of the product. In order to maximize the product yield, optimization of each process index is required, that is, a process index capable of maximizing the product yield is found. Therefore, at present, a relation model between the process index and the product yield is trained through machine learning, and then the process index capable of maximizing the process index is obtained on the model.
However, the process environment in the manufacturing industry is generally complex, so that a large amount of noise exists in the acquired index data of the process index, and the accuracy of the trained model is low.
Disclosure of Invention
in view of this, the present application provides a data processing method, an apparatus and an electronic device, so as to improve the accuracy of training a relationship model.
The application provides a data processing method, which comprises the following steps:
obtaining at least one training sample, wherein the training sample comprises index data of at least one index, the index data comprises sample data of the index of the training sample at least one moment, and the training sample has a preset yield label;
Denoising the sample data in the training sample based on the time attribute and the index attribute corresponding to the sample data;
training the initial model based on the denoised training sample and the yield label to obtain a relation model, wherein the relation model represents the relation between the index and the yield.
Optionally, the denoising method for the sample data in the training sample based on the time attribute and the index attribute corresponding to the sample data includes:
obtaining an incidence relation between the sample data in the index data of the target index; the target index is any one of the at least one index;
and calculating sample data in the index data of the target index based on the incidence relation to obtain new sample data of the target index at the moment.
Optionally, in the above method, the association relationship includes a similarity value between the sample data in the index data of the target index;
calculating sample data in the index data of the target index based on the incidence relation to obtain new sample data of the target index at the moment, wherein the method comprises the following steps:
Determining a sample weight coefficient corresponding to the time in the at least one time in the index data of the target index based on the similarity value;
and calculating sample data in the index data of the target index based on the sample weight coefficient to obtain new sample data of the target index at the moment.
optionally, the denoising method for the sample data in the training sample based on the time attribute and the index attribute corresponding to the sample data includes:
And updating the sample data based on the incidence relation between the time attribute and the index attribute corresponding to the sample data so as to realize the de-noising of the sample data in the training sample.
optionally, the updating the sample data by using the association relationship between the time attribute and the index attribute corresponding to the sample data includes:
and updating the sample data based on the incidence relation between the time attribute and the index attribute corresponding to the sample data by using an automatic attention mechanism.
optionally, in the method, the initial model is a fitting model built based on a preset machine learning algorithm, and the initial model has at least one initial fitting parameter related to the index and the yield.
The present application also provides a data processing apparatus, including:
the device comprises an obtaining unit, a calculating unit and a processing unit, wherein the obtaining unit is used for obtaining at least one training sample, the training sample comprises index data of at least one index, the index data comprises sample data of the index at least one moment, and the training sample has a preset yield label;
the denoising unit is used for denoising the sample data in the training sample based on the time attribute and the index attribute corresponding to the sample data;
and the training unit is used for training the initial model based on the denoised training sample and the yield label of the denoised training sample to obtain a relation model, and the relation model represents the relation between the index and the yield.
The above apparatus, optionally, the denoising unit includes:
The association obtaining subunit is configured to obtain an association relationship between the sample data in the index data of the target index; the target index is any one of the at least one index;
And the sample calculation subunit is used for calculating sample data in the index data of the target index based on the incidence relation so as to obtain new sample data of the target index at the moment.
The above apparatus, optionally, further comprises:
And the model building unit is used for building the initial model in advance based on a preset machine learning algorithm, the initial model is a fitting model, and the initial model is provided with at least one initial fitting parameter related to indexes and yield.
The present application further provides an electronic device, including:
the memory is used for storing the application program and data generated by the running of the application program;
A processor for executing the application to implement:
obtaining at least one training sample, wherein the training sample comprises index data of at least one index, the index data comprises sample data of the index of the training sample at least one moment, and the training sample has a preset yield label;
denoising the sample data in the training sample based on the time attribute and the index attribute corresponding to the sample data;
training the initial model based on the denoised training sample and the yield label to obtain a relation model, wherein the relation model represents the relation between the index and the yield.
According to the technical scheme, after the sample data of each index at each moment is obtained, the sample data is denoised based on the moment attribute and the index attribute in the sample data, and then the original model is trained by utilizing the training sample after the sample data is dried and the yield label of the training sample, so that the relation model between the index and the yield is obtained. Therefore, the sample data is denoised by the time attribute and the index attribute in the sample data, so that the sample data used for training the relational model is more accurate, the relational model trained by the sample data is more accurate, and the aim of improving the accuracy of the relational model is fulfilled.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.
fig. 1 is a flowchart of a data processing method according to an embodiment of the present application;
FIGS. 2 and 3 are schematic diagrams of embodiments of the present application;
Fig. 4 is a schematic structural diagram of a data processing apparatus according to a second embodiment of the present application;
FIG. 5 is a schematic partial structural diagram of a second embodiment of the present application;
FIG. 6 is another schematic structural diagram of a second embodiment of the present application;
Fig. 7 is a schematic structural diagram of an electronic device according to a third embodiment of the present application;
fig. 8 and 9 are diagrams illustrating an application example of the embodiment of the present application.
Detailed Description
the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
Referring to fig. 1, a flowchart of a data processing method provided in an embodiment of the present application is shown, where the method is applied to an electronic device having data processing capability and capable of performing model training, such as a terminal device like a computer or a server. The method is mainly used for denoising the sample data of the relation model between the training index and the yield, so that the aim of improving the accuracy of the relation model is fulfilled.
Specifically, the method in this embodiment may include the following steps:
step 101: at least one training sample is obtained.
the training sample comprises index data of at least one index. The index data includes sample data of the index to which the index belongs at least one time, that is, the training sample may be composed of at least one sample data, and may also include other sample data, where the sample data is sample data at least one time on at least one index, or may be expressed as: the sample data is sample data on at least one index at least one time instant.
It should be noted that the index may be a product production index, such as a temperature index, a flow index, or a pressure index, for a certain product in the manufacturing industry, and each index may have different sampling data at different times.
as shown in fig. 2, the training sample is composed of index data of 3 indexes: y1, y2, y3, the index data under each index is composed of sample data at 3 moments under the index: sample data for index 1 at 3 moments: y11, y12, y13, sample data for index 2 at 3 times: y21, y22, y23, sample data for index 3 at 3 times: y31, y32, y 33. Of course the training sample may also take another expression: the training sample consists of time data of 3 moments, and the time data at each moment consists of sample data on 3 indexes at the moment: sample data on 3 indices at time 1: y11, y21, y31, sample data on 3 indices at time 2: y12, y22, y32, sample data on 3 indices at time 3: y13, y23, y 33.
Specifically, in this embodiment, when obtaining the training sample, data at multiple times may be obtained as sample data for multiple indexes, respectively, and then the sample data may be combined to form one (set of) training sample, or data at multiple indexes may be obtained as sample data at multiple times, and then the sample data may be combined to form one (set of) training sample. In this embodiment, one or more (sets of) training samples may be obtained in this manner.
The training sample is provided with a preset yield label, wherein the yield label indicates a corresponding actual yield value under the sample data. Specifically, in this embodiment, the actual yield value under the sample data may be calculated, and the calculated actual yield value may be used as the yield label, or in this embodiment, the corresponding actual yield value collected in the history record for the sample data may be collected and used as the yield label, and so on.
Step 102: and denoising the sample data in the training sample based on the time attribute and the index attribute corresponding to the sample data.
the sample data under the same index attribute in the sample data may have a difference due to a difference in time attribute, so that denoising the training sample in this embodiment is implemented based on the time attribute and the index attribute corresponding to the sample data, which can be specifically understood as: in the embodiment, on the basis of the index attribute of the sample data, the sample data is denoised by combining the difference of the sample data in the time attribute; or, it can be understood that, in this embodiment, on the basis of the time attribute of the sample data, the sample data is denoised in combination with the association between the index attributes of the sample data.
For example, in this embodiment, the sample data with noise in the sample data is denoised in two dimensions, namely, a time attribute and an index attribute; or, in this embodiment, the sample data with noise in the sample data may be denoised according to the mutual influence between the time attribute and the index attribute on the sample data; or, in this embodiment, denoising sample data with noise in the sample data may be performed according to the association relationship between the time attribute and the index attribute, and so on.
Step 103: and training the initial model based on the denoised training sample and the yield label of the denoised training sample to obtain a relational model.
Wherein, the relationship model represents the corresponding relationship between the index and the yield.
specifically, in this embodiment, the initial model is trained by using the denoised training sample and the yield label thereof, and mainly, various model parameters in the initial model are continuously optimized until the model parameters are optimal, and the training is ended, at this time, the initial model forms a relationship model for completing the training based on the optimized model parameters. For example, in this embodiment, the model parameters of the initial model are optimized by using the training samples and the yield labels in an iterative manner, and in each iterative training, whether the model parameters converge or not, that is, whether the model parameters tend to be stable or not, is monitored until the model parameters are stable along with multiple times of training, that is, the parameter results of the two previous and subsequent training do not change, at this time, the model parameters are determined to be the optimal model parameters, and thus, a trained relational model is obtained, and the relational model can be used for optimizing the index for maximizing the product yield.
According to the scheme, after the sample data of each index at each time is obtained, the sample data is denoised based on the time attribute and the index attribute of the sample data, and the initial model is trained by using the training sample after the data is dried and the yield label of the training sample, so that the relation model between the index and the yield is obtained. Therefore, in the embodiment, the time attribute and the index attribute in the sample data are used for denoising the sample data, so that the sample data used for training the relational model is more accurate, the relational model trained by using the sample data is more accurate, and the purpose of improving the accuracy of the relational model is achieved.
in an implementation manner, when denoising the sample data, step 102 in this embodiment may be specifically implemented by:
Firstly, obtaining the incidence relation among sample data in the index data of the target index, and respectively calculating the sample data in the index data of the target index based on the incidence relation to obtain new sample data of the target index at the moment.
The target index is any one of at least one index in the training samples, that is, the following operations are performed on any one index in the training samples in this embodiment:
And obtaining the sample data of the target index at each moment in at least one moment, obtaining the incidence relation among the sample data, namely obtaining the incidence relation among the sample data at each moment on the target index, and recalculating the sample data at each moment on the target index based on the incidence relation to obtain new sample data at each moment on the target index.
Examples are as follows: for sample data of 3 indexes in a training sample at 3 moments: y11, y12, y13, y21, y22, y23, y31, y32, y33, the following operations are respectively executed for the sample data of each index at 3 moments:
Respectively calculating the association relations among y11, y12 and y13 of sample data y11, y12 and y13 of the index 1 at 3 moments, and then respectively recalculating y11, y12 and y13 based on the association relations among y11, y12 and y13 to obtain new sample data y11 ', y12 ' and y13 ' of the index 1 at 3 moments;
Respectively calculating the association relations among y21, y22 and y23 of sample data y21, y22 and y23 of the index 2 at 3 moments, and then respectively recalculating y21, y22 and y23 based on the association relations among y21, y22 and y23 to obtain new sample data y21 ', y22 ' and y23 ' of the index 2 at 3 moments;
And respectively calculating the association relations among y31, y32 and y33 of sample data y31, y32 and y33 of the index 3 at 3 moments, and then respectively recalculating y31, y32 and y33 based on the association relations among y31, y32 and y33 to obtain new sample data y31 ', y32 ' and y33 ' of the index 3 at 3 moments.
specifically, the association relationship between the sample data in this embodiment may be a similarity value between sample data in the index data of the target index. Such as similarity values between y31, y32, y33, and the like.
Correspondingly, when the sample data is calculated based on the incidence relation to obtain new sample data, the method can be realized in the following mode:
first, based on the similarity value, a sample weight coefficient corresponding to a time instant of at least one time instant in the index data of the target index is determined, that is, based on the similarity value, a sample weight coefficient of sample data at each time instant on the target index is determined, as for y11, the sample weight coefficient thereof includes: for y11, y12, and y13, respectively: 0.1, 0.3, 0.6; for y12, the sample weight coefficients include: for y11, y12, and y13, respectively: 0.2, 0.4, for y13, the sample weight coefficients include: for y11, y12, and y13, respectively: 0.3, 0.4, 0.3;
Then, based on the sample weight coefficient, sample data in the index data of the target index is calculated to obtain new sample data of the target index at that time, that is, each sample data in the index data of the target index is recalculated by using the sample weight coefficient, so as to obtain new sample data at each time.
for example, after a sample weight coefficient of sample data at each time is obtained, the sample weight coefficient of the sample data is used to perform calculation in combination with other sample data of the same target index as the sample data, thereby obtaining new sample data.
Examples are as follows: as for y11, the sample weight coefficients for y11 include: for y11, y12, and y13, respectively: 0.1, 0.3, 0.6, the value obtained by 0.1 y11+0.3 y12+0.6 y13 was taken as new y 11'; similarly, for y12, the y12 sample weight coefficients include: for y11, y12, and y13, respectively: 0.2, 0.4, the value obtained by 0.2 y11+0.4 y12+0.4 y13 was taken as new y 12'; for y13, the y13 sample weight coefficients include: for y11, y12, and y13, respectively: 0.3, 0.4, 0.3, the values obtained from 0.3 y11+0.4 y12+0.3 y13 were taken as new y 13' as shown in fig. 3.
It can be seen that, in the new sample data obtained in this embodiment, first, a plurality of sample data are integrated to calculate correlations, such as similarity values, between the plurality of sample data, and then, each sample data is regenerated by using the similarity values between the plurality of sample data, so that data between indexes and between samples are effectively used, noise data can be effectively reduced, and accuracy of the trained relationship model is improved.
It should be noted that, in this embodiment, the above denoising of the sample data can be realized by a self-attention mechanism (self-attention).
In another implementation manner, when denoising the sample data, the step 102 in this embodiment may be specifically implemented by:
And updating the sample data based on the incidence relation between the time attribute and the index attribute corresponding to the sample data so as to realize the denoising of the sample data in the training sample.
The sample data has a certain incidence relation between the time attribute and the index attribute, for example, the sample data corresponding to different time attributes on the same index attribute has similarity or presents a change relation with a certain rule, such as increasing or transferring or changing continuously within a certain amplitude range; for another example, the sample data on different index attributes corresponding to the same time attribute has a certain corresponding relationship, such as a relationship of eliminating the length of the sample data, a linear increasing or decreasing relationship, a nonlinear relationship or an exponential relationship, and the like, so in this embodiment, the sample data is updated in combination with the association relationship between the time attribute and the index attribute corresponding to the sample data, for example, the sample data is modified (increased or decreased, and the like) or new sample data is regenerated, thereby implementing the denoising processing on the sample data.
In a specific implementation, when denoising is performed, a preset denoising mechanism may be adopted in this embodiment, for example, a self-attention mechanism is used, and sample data is updated based on an association relationship between a time attribute and an index attribute corresponding to the sample data.
For example, a self-attention mechanism is utilized to expand the time dimension of the sample data on each index, and the sample data on more times on each index is expanded according to the incidence relation between the time attribute and the index attribute corresponding to the sample data, so that the sample data is updated in a data expansion mode, the sample data is richer, the proportion occupied by the sample data with noise is smaller at the level of big data, and the denoising of the sample data is realized;
Or, the sample data is modified on the time dimension of each index by using a self-attention mechanism, specifically, the sample data on each time on each index is modified according to the incidence relation between the time attribute corresponding to the sample data and the index attribute, for example, the sample data is increased or decreased according to a proportion, or the sample data is directly increased or decreased in value, and the like, so that the sample data is updated in a data modification manner, the sample data is more accurate, and the denoising of the sample data is realized, and the like.
Examples are as follows: for sample data of 3 indexes in a training sample at 3 moments: y11, y12, y13, y21, y22, y23, y31, y32 and y33, wherein a new representation is generated for the sample data by combining the incidence relation of the sample data between the time attribute and the index attribute, namely, new sample data, the sample data can be expanded to the sample data on m dimensions (time) on each index attribute, at the moment, m sample data are arranged on each index of the training sample on the 3 index attributes, and correspondingly, model training is carried out on the basis of the expanded sample data, so that the optimization of model parameters is realized, a relation model representing the relation between the index and the yield is obtained, and the method can be used for optimizing the index with the maximized yield.
therefore, the new sample data obtained in the embodiment integrates the association between the index and the two dimensions of the time, and the sample data is regenerated, so that the association between the index and the association between the samples are effectively utilized, the noise data can be effectively reduced, and the accuracy of the trained relation model is improved.
based on the implementation, the initial model in this embodiment may be a fitting model constructed based on a preset machine learning algorithm, the fitting model has at least one initial fitting parameter, and the training sample after denoising and the yield label thereof are utilized to train and optimize the initial fitting parameters in this embodiment, so that the trained relation fitting model is more accurate, and the index is more accurately found when the yield is maximized.
The machine learning algorithm may be a neural network learning algorithm or a convolutional network algorithm.
Referring to fig. 4, a schematic structural diagram of a data processing apparatus provided in the second embodiment of the present application is shown, where the apparatus may be disposed in an electronic device, such as a computer or a server, which has data processing capability and is capable of performing model training. The device in the application is mainly used for denoising the sample data of the relation model between the training index and the yield, so that the aim of improving the accuracy of the relation model is fulfilled.
Specifically, the apparatus in this embodiment may include the following functional units:
an obtaining unit 401 for obtaining at least one training sample.
the training sample comprises index data of at least one index. The index data includes sample data of the index to which the index belongs at least one time, that is, the training sample is composed of at least one sample data, and the sample data is sample data at least one time on at least one index, or may be expressed as: the sample data is sample data on at least one index at least one time instant.
it should be noted that the index may be a product production index, such as a temperature index, a flow index, or a pressure index, for a certain product in the manufacturing industry, and each index may have different sampling data at different times.
as shown in fig. 2, the training sample is composed of index data of 3 indexes: y1, y2, y3, the index data under each index is composed of sample data at 3 moments under the index: sample data for index 1 at 3 moments: y11, y12, y13, sample data for index 2 at 3 times: y21, y22, y23, sample data for index 3 at 3 times: y31, y32, y 33. Of course the training sample may also take another expression: the training sample consists of time data of 3 moments, and the time data at each moment consists of sample data on 3 indexes at the moment: sample data on 3 indices at time 1: y11, y21, y31, sample data on 3 indices at time 2: y12, y22, y32, sample data on 3 indices at time 3: y13, y23, y 33.
Specifically, in this embodiment, when obtaining the training sample, data at multiple times may be obtained as sample data for multiple indexes, respectively, and then the sample data may be combined to form one (set of) training sample, or data at multiple indexes may be obtained as sample data at multiple times, and then the sample data may be combined to form one (set of) training sample. In this embodiment, one or more (sets of) training samples may be obtained in this manner.
The training sample is provided with a preset yield label, wherein the yield label indicates a corresponding actual yield value under the sample data. Specifically, in this embodiment, the actual yield value under the sample data may be calculated, and the calculated actual yield value may be used as the yield label, or in this embodiment, the corresponding actual yield value collected in the history record for the sample data may be collected and used as the yield label, and so on.
A denoising unit 402, configured to denoise the sample data in the training sample based on a time attribute and an index attribute corresponding to the sample data;
The sample data under the same index attribute in the sample data may have a difference due to a difference in time attribute, so that denoising the training sample in this embodiment is implemented based on the time attribute and the index attribute corresponding to the sample data, which can be specifically understood as: in the embodiment, on the basis of the index attribute of the sample data, the sample data is denoised by combining the difference of the sample data in the time attribute; or, it can be understood that, in this embodiment, on the basis of the time attribute of the sample data, the sample data is denoised in combination with the association between the index attributes of the sample data.
For example, in this embodiment, the sample data with noise in the sample data is denoised in two dimensions, namely, a time attribute and an index attribute; or, in this embodiment, the sample data with noise in the sample data may be denoised according to the mutual influence between the time attribute and the index attribute on the sample data; or, in this embodiment, denoising sample data with noise in the sample data may be performed according to the association relationship between the time attribute and the index attribute, and so on.
The training unit 403 is configured to train the initial model based on the denoised training sample and the yield label that the denoised training sample has, to obtain a relationship model.
wherein, the relationship model represents the corresponding relationship between the index and the yield.
Specifically, in this embodiment, the initial model is trained by using the denoised training sample and the yield label thereof, and mainly, various model parameters in the initial model are continuously optimized until the model parameters are optimal, and the training is ended, at this time, the initial model forms a relationship model for completing the training based on the optimized model parameters. For example, in this embodiment, the model parameters of the initial model are optimized by using the training samples and the yield labels in an iterative manner, and in each iterative training, whether the model parameters converge or not, that is, whether the model parameters tend to be stable or not, is monitored until the model parameters are stable along with multiple times of training, that is, the parameter results of the two previous and subsequent training do not change, at this time, the model parameters are determined to be the optimal model parameters, and thus, a trained relational model is obtained, and the relational model can be used for optimizing the index for maximizing the product yield.
As can be seen from the above solutions, in the data processing apparatus provided in the second embodiment of the present application, after sample data of each index at each time is obtained, based on the time attribute and the index attribute in the sample data, the sample data is denoised, and then the initial model is trained by using the training sample after being dried and the yield label of the training sample, so as to obtain the relationship model between the index and the yield. Therefore, in the embodiment, the time attribute and the index attribute in the sample data are used for denoising the sample data, so that the sample data used for training the relational model is more accurate, the relational model trained by using the sample data is more accurate, and the purpose of improving the accuracy of the relational model is achieved.
in one implementation, the denoising unit 402 may include the following structure therein, as shown in fig. 5:
an association obtaining subunit 421, configured to obtain an association relationship between the sample data in the index data of the target index; the target index is any one of the at least one index;
and the sample calculation subunit 422 is configured to calculate, based on the association relationship, sample data in the index data of the target index to obtain new sample data of the target index at the time.
In addition, the apparatus in this embodiment may further include the following structural units, as shown in fig. 6:
A model building unit 404, configured to build the initial model based on a preset machine learning algorithm in advance, where the initial model is a fitting model and has at least one initial fitting parameter related to an index and a yield.
it should be noted that, for the specific implementation of each unit in the apparatus of the present embodiment, reference may be made to the corresponding drawings and contents in the foregoing, and details are not described herein.
referring to fig. 7, a schematic structural diagram of an electronic device according to a third embodiment of the present disclosure is provided, where the electronic device may be a device having data processing capability and capable of performing model training, such as a terminal device like a computer or a server. The electronic equipment is mainly used for denoising the sample data of the relation model between the training index and the yield, so that the aim of improving the accuracy of the relation model is fulfilled.
Specifically, the electronic device in this embodiment may include the following structure:
A memory 701 for storing an application program and data generated by the application program;
a processor 702 for executing the application to implement:
Obtaining at least one training sample, wherein the training sample comprises index data of at least one index, the index data comprises sample data of the index of the training sample at least one moment, and the training sample has a preset yield label; denoising the sample data in the training sample based on the time attribute and the index attribute corresponding to the sample data; training the initial model based on the denoised training sample and the yield label to obtain a relation model, wherein the relation model represents the relation between the index and the yield.
according to the scheme, after the sample data of each index at each time is obtained, the electronic device provided by the third embodiment of the application denoises the sample data based on the time attribute and the index attribute in the sample data, and trains the initial model by using the training sample after the sample data is dried and the yield label of the training sample to obtain the relation model between the index and the yield. Therefore, in the embodiment, the time attribute and the index attribute in the sample data are used for denoising the sample data, so that the sample data used for training the relational model is more accurate, the relational model trained by using the sample data is more accurate, and the purpose of improving the accuracy of the relational model is achieved.
in one implementation, the processor 702 may perform denoising by:
obtaining an incidence relation between the sample data in the index data of the target index; the target index is any one of the at least one index;
and calculating sample data in the index data of the target index based on the incidence relation to obtain new sample data of the target index at the moment.
Wherein the incidence relation comprises a similarity value between the sample data in the index data of the target index; correspondingly, based on the association relationship, sample data in the index data of the target index is calculated to obtain new sample data of the target index at the time, which can be realized by the following method:
Determining a sample weight coefficient corresponding to the time in the at least one time in the index data of the target index based on the similarity value; and calculating sample data in the index data of the target index based on the sample weight coefficient to obtain new sample data of the target index at the moment.
in one implementation, the processor 702 may also perform denoising by:
and updating the sample data based on the incidence relation between the time attribute and the index attribute corresponding to the sample data so as to realize the de-noising of the sample data in the training sample.
For example, the sample data is updated based on the association relationship between the time attribute and the index attribute corresponding to the sample data by using the self-attention mechanism.
in addition, the initial model is a fitting model built based on a preset machine learning algorithm, and the initial model is provided with at least one initial fitting parameter related to indexes and yield
It should be noted that, for the implementation of the processor in the electronic device of the present embodiment, reference is made to the corresponding drawings and contents in the foregoing, and details are not described here.
In the embodiment, a noise data processing mechanism based on self-attention is added to a fitting model used for optimizing the index when the yield is maximized, so that a training sample in model training is denoised firstly, and then model parameters are optimized and trained, so that a more accurate fitting model is obtained for index optimization. The core process in this embodiment is described below:
1) Data acquisition and sample construction, i.e. acquiring training samples, including sampled data (sample data) at multiple times on multiple indicators
2) fitting model design with self-attention noise data processing mechanism
Among these, due to noisy plant environment, the above data used to construct sample data usually contains noisy data, and the constructed sample data also contains relatively large amounts of noisy data, which adversely affects the construction of a yield model (i.e., a relational model).
Therefore, in the embodiment, model training is performed after the sample data is denoised, so that the accuracy of the model is improved. Specifically, in this embodiment, a self-annotation structure is used to combine the relevant information of the sample data in two latitudes, namely the time sequence (time) and the characteristic (index), to generate a new representation (new sample data) for the sample data of the incoming model, which effectively reduces the noise in the original sample data.
3) Model training
and (3) substituting the sample constructed in the step (1) into the fitting model with self-attention noise processing mechanism designed in the step (2) for training to generate a fitting relation model showing the relation between the index and the yield.
Therefore, the technical scheme in the embodiment has the following advantages:
1) Denoising the sample data by using a self-attention (self-attention) mechanism, wherein the self-attention mechanism can be combined with a time sequence and two latitudes of an index to perform denoising processing such as smoothing, expansion or modification on the data, and more information is referred to during denoising processing, so that the noise processing effect is better;
2) the self-attention (self-attention) mechanism is embedded into the model training process as part of the whole model training, and provides an end-to-end complete process from noise data processing to model training. The problem of poor noise data processing effect caused by discontinuous noise data processing and model training in the traditional noise data processing scheme is solved.
in a specific implementation, as shown by the overall framework diagram shown in fig. 8 and the schematic diagram of the fitting model training logic with self-attention noise data processing mechanism shown in fig. 9, the present embodiment mainly includes three stages of "data preparation", "model design" and "model training" when modeling the high-value product yield in the industrial field using the self-attention (self-attention) mechanism to perform noise data processing, where:
1 data preparation
The method mainly comprises the steps of acquiring indexes (represented by X1, X2, device temperature and the like) which reflect the operation conditions of a product production device and actual high-value product yield (represented by Y, such as the gasoline yield in the petrochemical industry) and collected by a database (such as an IP21 real-time database) of a target process manufacturing enterprise which needs index optimization of high-value product yield maximization, and constructing the data into sample data. As shown in the sample data of table 1.
TABLE 1 sample data
time index | X1 | X2 | ... | Xn | yield Y |
T1 | x1_1 | X2_1 | ... | Xn_1 | Y_1 |
T2 | X1_2 | X2_2 | ... | Xn_2 | Y_2 |
T3 | X1_3 | X2_3 | ... | Xn_3 | Y_3 |
T4 | X1_4 | X2_4 | ... | Xn_4 | Y_4 |
... | ... | ... | ... | ... | ... |
TM | X1_m | X2_m | ... | Xn_m | Y_m |
Where "time" is the time of acquisition of the indicator, the high value product yield Y can be obtained directly by a meter on the product production device (if there is a corresponding measurement point on the device) or indirectly calculated by the values of other measurement points.
2 constructing a yield fitting model with self-attention noise processing mechanism
as shown in the logical hierarchy of model training in fig. 9:
the input layer indicates input of sample data corresponding to each time under each index. For example, three indices X1, X2, X3 represent combinations of samples at three consecutive time instants, and the respective three dimensions (e.g., X1: X11, X12, X13) represent different indices for the samples at each time instant, the actual number being related to the number of selected plant process indices, and there may be other numbers of time instants;
And a self-attention (self-attention) layer for calculating correlations among samples (corresponding to X1, X2 and X3) transmitted from the input layer, and regenerating a vector representation (new sample data) corresponding to each sample based on the correlations among the samples such as similarity values, for example, such that the representation of X1 is re-represented as [ Xs11, Xs12, … and Xs1m ] from the original representation of [ X11, X12 and X13 ]. For example, a new vector representation of the sample data is implemented by using the technique of setting a weight coefficient by a similarity value and performing the sample data as described above. The vector representation firstly integrates the information of a plurality of characteristics to calculate the relevance among a plurality of samples, and then the vector representation of each sample is regenerated by utilizing the relevance among a plurality of samples. Effectively utilizing the information between the characteristics and the information between the samples, effectively reducing noise data and improving the accuracy of the fitting model.
a hidden layer which is a conventional fully-connected layer, is represented by a vector generated by a self-attention (self-attention) mechanism in a front-end mode, is represented by a back-end output layer, and outputs intermediate quantities H11-H1n for the output layer;
and the output layer is a fully-connected layer with sigmoid activation, and the yield Y is output on H.
it should be noted that fig. 9 is only a general structure of the yield model, and specific implementations may have different manners, such as the number of self-attention (self-attention) layers and hidden layers, and the number of neurons in each layer may be adjusted according to actual situations, and is not unique.
3 model training
And on the basis of obtaining sample data and constructing a model, sending the data prepared in the step 1 into the model designed in the step 2 for training. And adjusting and optimizing model parameters of the model in the training process to obtain a parameter combination capable of performing accurate fitting to serve as a final fitting model.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description.
Those of skill would further appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that the various illustrative components and steps have been described above generally in terms of their functionality in order to clearly illustrate this interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module may reside in Random Access Memory (RAM), memory, Read Only Memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (10)
1. A method of data processing, comprising:
Obtaining at least one training sample, wherein the training sample comprises index data of at least one index, the index data comprises sample data of the index of the training sample at least one moment, and the training sample has a preset yield label;
denoising the sample data in the training sample based on the time attribute and the index attribute corresponding to the sample data;
Training the initial model based on the denoised training sample and the yield label to obtain a relation model, wherein the relation model represents the relation between the index and the yield.
2. The method of claim 1, denoising the sample data in the training sample based on the time attribute and the index attribute corresponding to the sample data, comprising:
obtaining an incidence relation between the sample data in the index data of the target index; the target index is any one of the at least one index;
And calculating sample data in the index data of the target index based on the incidence relation to obtain new sample data of the target index at the moment.
3. The method of claim 2, the correlation comprising a similarity value between the sample data in the target indicator's indicator data;
Calculating sample data in the index data of the target index based on the incidence relation to obtain new sample data of the target index at the moment, wherein the method comprises the following steps:
determining a sample weight coefficient corresponding to the time in the at least one time in the index data of the target index based on the similarity value;
And calculating sample data in the index data of the target index based on the sample weight coefficient to obtain new sample data of the target index at the moment.
4. the method of claim 1, denoising the sample data in the training sample based on the time attribute and the index attribute corresponding to the sample data, comprising:
and updating the sample data based on the incidence relation between the time attribute and the index attribute corresponding to the sample data so as to realize the de-noising of the sample data in the training sample.
5. The method according to claim 4, updating the sample data by using the association relationship between the time attribute and the index attribute corresponding to the sample data, comprising:
And updating the sample data based on the incidence relation between the time attribute and the index attribute corresponding to the sample data by using an automatic attention mechanism.
6. the method according to claim 1, 2 or 4, wherein the initial model is a fitting model constructed based on a preset machine learning algorithm, and the initial model has at least one initial fitting parameter related to indexes and yield.
7. a data processing apparatus comprising:
the device comprises an obtaining unit, a calculating unit and a processing unit, wherein the obtaining unit is used for obtaining at least one training sample, the training sample comprises index data of at least one index, the index data comprises sample data of the index at least one moment, and the training sample has a preset yield label;
The denoising unit is used for denoising the sample data in the training sample based on the time attribute and the index attribute corresponding to the sample data;
And the training unit is used for training the initial model based on the denoised training sample and the yield label of the denoised training sample to obtain a relation model, and the relation model represents the relation between the index and the yield.
8. The apparatus of claim 7, the denoising unit, comprising:
The association obtaining subunit is configured to obtain an association relationship between the sample data in the index data of the target index; the target index is any one of the at least one index;
And the sample calculation subunit is used for calculating sample data in the index data of the target index based on the incidence relation so as to obtain new sample data of the target index at the moment.
9. the apparatus of claim 7 or 8, further comprising:
and the model building unit is used for building the initial model in advance based on a preset machine learning algorithm, the initial model is a fitting model, and the initial model is provided with at least one initial fitting parameter related to indexes and yield.
10. an electronic device, comprising:
The memory is used for storing the application program and data generated by the running of the application program;
A processor for executing the application to implement:
obtaining at least one training sample, wherein the training sample comprises index data of at least one index, the index data comprises sample data of the index of the training sample at least one moment, and the training sample has a preset yield label;
denoising the sample data in the training sample based on the time attribute and the index attribute corresponding to the sample data;
Training the initial model based on the denoised training sample and the yield label to obtain a relation model, wherein the relation model represents the relation between the index and the yield.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910848176.5A CN110569966B (en) | 2019-09-09 | 2019-09-09 | Data processing method and device and electronic equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910848176.5A CN110569966B (en) | 2019-09-09 | 2019-09-09 | Data processing method and device and electronic equipment |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110569966A true CN110569966A (en) | 2019-12-13 |
CN110569966B CN110569966B (en) | 2022-09-23 |
Family
ID=68778564
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910848176.5A Active CN110569966B (en) | 2019-09-09 | 2019-09-09 | Data processing method and device and electronic equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110569966B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111340276A (en) * | 2020-02-19 | 2020-06-26 | 联想(北京)有限公司 | Method and system for generating prediction data |
CN112711578A (en) * | 2020-12-30 | 2021-04-27 | 陈静 | Big data denoising method for cloud computing service and cloud computing financial server |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TW200540674A (en) * | 2004-06-03 | 2005-12-16 | Univ Nat Cheng Kung | Quality prognostics system and method for manufacturing processes |
CN106249724A (en) * | 2016-09-14 | 2016-12-21 | 东北大学 | A kind of blast furnace polynary molten steel quality forecast Control Algorithm and system |
CN109636014A (en) * | 2018-11-28 | 2019-04-16 | 郑州轻工业学院 | A kind of method of distributing cotton based on yarn quality prediction |
-
2019
- 2019-09-09 CN CN201910848176.5A patent/CN110569966B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TW200540674A (en) * | 2004-06-03 | 2005-12-16 | Univ Nat Cheng Kung | Quality prognostics system and method for manufacturing processes |
CN106249724A (en) * | 2016-09-14 | 2016-12-21 | 东北大学 | A kind of blast furnace polynary molten steel quality forecast Control Algorithm and system |
CN109636014A (en) * | 2018-11-28 | 2019-04-16 | 郑州轻工业学院 | A kind of method of distributing cotton based on yarn quality prediction |
Non-Patent Citations (2)
Title |
---|
方积乾: "《医学统计手册》", 31 May 2018 * |
田翠华: "《基于GT4的物联网交通信息服务仿真研究》", 31 January 2017 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111340276A (en) * | 2020-02-19 | 2020-06-26 | 联想(北京)有限公司 | Method and system for generating prediction data |
CN111340276B (en) * | 2020-02-19 | 2022-08-19 | 联想(北京)有限公司 | Method and system for generating prediction data |
CN112711578A (en) * | 2020-12-30 | 2021-04-27 | 陈静 | Big data denoising method for cloud computing service and cloud computing financial server |
CN112711578B (en) * | 2020-12-30 | 2021-09-21 | 深圳市全景网络有限公司 | Big data denoising method for cloud computing service and cloud computing financial server |
Also Published As
Publication number | Publication date |
---|---|
CN110569966B (en) | 2022-09-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110674604B (en) | Transformer DGA data prediction method based on multi-dimensional time sequence frame convolution LSTM | |
CN109492808B (en) | Method for predicting remaining parking spaces of indoor parking lot | |
CN111860982A (en) | Wind power plant short-term wind power prediction method based on VMD-FCM-GRU | |
CN106487359B (en) | The building method of Morphologic filters based on self-adapting multi-dimension AVG-Hat transformation | |
CN110569966B (en) | Data processing method and device and electronic equipment | |
CN109635010B (en) | User characteristic and characteristic factor extraction and query method and system | |
EP4280088A1 (en) | Time series data trend feature extraction method based on dynamic grid division | |
CN113609932B (en) | Fault detection method, system, device and medium based on long-term and short-term memory network | |
CN113779882A (en) | Method, device, equipment and storage medium for predicting residual service life of equipment | |
CN114548199A (en) | Multi-sensor data fusion method based on deep migration network | |
CN108647714A (en) | Acquisition methods, terminal device and the medium of negative label weight | |
CN117786371B (en) | Temperature monitoring data optimization prediction analysis method and system | |
CN113670609B (en) | Fault detection method, system, device and medium based on wolf optimization algorithm | |
CN109255368B (en) | Method, device, electronic equipment and storage medium for randomly selecting characteristics | |
CN114978931B (en) | Network traffic prediction method and device based on manifold learning and storage medium | |
CN116883065A (en) | Merchant risk prediction method and device | |
CN117093830A (en) | User load data restoration method considering local and global | |
CN115836849A (en) | Arrhythmia classification method based on comparison test time adaptation | |
CN115035304A (en) | Image description generation method and system based on course learning | |
CN113609933B (en) | Fault detection method, system, device and storage medium based on suffix tree | |
CN116975517B (en) | Sparse recovery method and system for partial weighted random selection strategy | |
CN116843368B (en) | Marketing data processing method based on ARMA model | |
Zong et al. | Research on the Prediction Model of House Rent Based on Machine Learning | |
CN115859201A (en) | Chemical process fault diagnosis method and system | |
Mody et al. | Enhancing Real Estate Market Insights through Machine Learning: Predicting Property Prices with Advanced Data Analytics |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |