CN113452379B - Section contour dimension reduction model training method and system and data compression method and system - Google Patents
Section contour dimension reduction model training method and system and data compression method and system Download PDFInfo
- Publication number
- CN113452379B CN113452379B CN202110805792.XA CN202110805792A CN113452379B CN 113452379 B CN113452379 B CN 113452379B CN 202110805792 A CN202110805792 A CN 202110805792A CN 113452379 B CN113452379 B CN 113452379B
- Authority
- CN
- China
- Prior art keywords
- model
- training
- dimension reduction
- reduction model
- section contour
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000009467 reduction Effects 0.000 title claims abstract description 111
- 238000012549 training Methods 0.000 title claims abstract description 86
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000013144 data compression Methods 0.000 title claims description 15
- 238000005457 optimization Methods 0.000 claims abstract description 16
- 238000004519 manufacturing process Methods 0.000 claims abstract description 12
- 238000013528 artificial neural network Methods 0.000 claims abstract description 11
- 238000012360 testing method Methods 0.000 claims description 22
- 230000006870 function Effects 0.000 claims description 15
- 210000002569 neuron Anatomy 0.000 claims description 10
- 239000000463 material Substances 0.000 claims description 8
- 238000012795 verification Methods 0.000 claims description 8
- 238000010276 construction Methods 0.000 claims description 2
- 238000007906 compression Methods 0.000 abstract description 13
- 230000006835 compression Effects 0.000 abstract description 13
- 230000005540 biological transmission Effects 0.000 abstract description 4
- 238000003860 storage Methods 0.000 abstract description 4
- 238000005096 rolling process Methods 0.000 description 7
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical compound [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 229910000831 Steel Inorganic materials 0.000 description 3
- 230000004913 activation Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 239000010959 steel Substances 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 229910052742 iron Inorganic materials 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 241001270131 Agaricus moelleri Species 0.000 description 1
- 101000608720 Helianthus annuus 10 kDa late embryogenesis abundant protein Proteins 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
- H03M7/00—Conversion of a code where information is represented by a given sequence or number of digits to a code where the same, similar or subset of information is represented by a different sequence or number of digits
- H03M7/30—Compression; Expansion; Suppression of unnecessary data, e.g. redundancy reduction
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The invention relates to a method and a system for training a section profile dimension reduction model, wherein the initial section profile dimension reduction model is constructed based on a noise reduction self-coding neural network, and the trained section profile dimension reduction model is obtained through optimization training, so that the dimension reduction model which accords with the actual production process of strip section profile data can be obtained. The invention also relates to a compression method of the strip section profile data, which compresses the section profile data in the production process of the strip by using the section profile dimensionality reduction model, meets the real-time compression requirement of the strip section profile data and is convenient for storage and transmission of the strip section profile data.
Description
Technical Field
The invention relates to the technical field of metallurgical rolling, in particular to a cross section profile dimensionality reduction model training method and system and a data compression method and system.
Background
With the rapid development of modern industry, the steel industry in China is transforming from the big iron and steel countries to the strong iron and steel countries, and the demand of users on high-quality strips is increasing day by day. The section shape is one of important quality indexes of strip products, and in order to improve the product quality, production enterprises often configure a convexity meter in a machine set to record and store strip profile data in real time so as to effectively control and trace back and detect the section shape of the strip in a later period. The profile data is a set of measurement data of the convexity meter on the strip thickness along the strip width direction, the dimensionality of the profile data is related to the strip width and the number of detection channels of the convexity meter, generally between 20-40, along with the rolling process, each roll of strip can generate hundreds of thousands of sets of profile data along the rolling direction, as the production is carried out continuously, a large amount of profile data can be generated, and meanwhile, the data can be stored in a production field only within a certain time, and meanwhile, the data volume is too large to be beneficial to remote transmission.
Therefore, a data compression method aiming at the strip section profile data needs to be established.
Disclosure of Invention
The invention aims to provide a cross section profile dimensionality reduction model training method and a cross section profile dimensionality reduction model training system, which can perform dimensionality reduction training on a large amount of cross section profile data by introducing a noise reduction self-coding neural network model.
The invention also provides a data compression method and a data compression system, which realize the compression of the strip section outline data through the section outline dimension reduction model, reduce the data volume and ensure the accuracy of the data.
In order to achieve the purpose, the invention provides the following scheme:
a cross section contour dimension reduction model training method comprises the following steps:
constructing an initial section contour dimension reduction model; the section contour dimension reduction model is constructed based on a self-coding neural network;
and carrying out optimization training on the initial section contour dimension reduction model to obtain a trained section contour dimension reduction model.
The invention also provides a cross section contour dimension reduction model training system, which comprises:
the model construction module is used for constructing an initial section contour dimension reduction model;
and the optimization training module is used for performing optimization training on the initial section contour dimension reduction model to obtain a trained section contour dimension reduction model.
The invention also provides a strip section contour data compression method, which comprises the following steps:
and compressing the strip section profile data by using the section profile dimension reduction model.
The invention also provides a strip section profile data compression system, which comprises:
and the data compression module is used for compressing the strip material section profile data by using the section profile dimension reduction model.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
according to the cross section profile dimensionality reduction model training method and system, and the data compression method and system, the cross section profile dimensionality reduction model is constructed by introducing the noise reduction self-coding neural network, the dimensionality reduction training can be carried out on a large amount of cross section profile data, the optimal compression dimensionality with the minimum distortion is obtained, real information is reserved to the greatest extent, the execution speed is high, the requirement for online use can be met, the storage and transmission efficiency of the strip cross section profile data is effectively improved, the monitoring and the tracing of strip product information are facilitated, and the important significance is achieved for improving the strip product quality.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.
FIG. 1 is a flowchart of a cross-sectional profile dimension reduction model training method according to an embodiment of the present invention;
FIG. 2 is a diagram of a training error loss of a cross-sectional profile dimensionality reduction model provided by an embodiment of the present invention;
FIG. 3 is a comparison graph of the original value of the profile of the cross section of the strip and the reconstructed value of the model provided by the embodiment of the invention;
FIG. 4(a) is a gray scale of the original value of the cross-sectional profile of the strip provided by the embodiment of the present invention, and (b) is a gray scale of the reconstructed value of the cross-sectional profile of the strip provided by the embodiment of the present invention;
FIG. 5 is a schematic structural diagram of a noise-reducing self-coding neural network according to an embodiment of the present invention;
fig. 6 is a block diagram of a cross-sectional profile dimension reduction model training system according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, 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 invention.
At present, with the rapid development of network technology and data acquisition and storage technology, data characteristics are becoming diversified. When large data is processed, not only are more strict requirements on acquisition, storage, bandwidth and the like provided, but also severe challenges are brought to tasks such as data analysis, compression, reconstruction and the like. Some theoretical and technical studies have been made on data compression, and image compression is more typical. For example, when a hyperspectral image is compressed by a tensor compression algorithm based on sparse representation, the method is superior to other algorithms in the aspects of original data information retention and denoising capability. The remote sensing image compression method based on the depth self-coding network is combined with the network structure characteristics of CNN and BNN, and a convolution self-coding and decoding network compression model added with binary noise is set up to be better than a traditional compression model in compression of the remote sensing image, and the image reconstruction effect is improved. The data dimension reduction method based on the sparse self-coding network abstracts and expresses face image data through a sparse coding theory, removes redundant information of images, reduces dimensions of the data, simultaneously retains main characteristics of faces, and obtains an identification rate close to 95%.
Therefore, the invention aims to provide a method and a system for training a section profile data dimension reduction model, a method and a system for data compression, which can well compress the dimension of the section profile data in the width direction of a strip material, greatly reduce the data volume, and have small error and low distortion rate between the section profile data after model reconstruction and the real section profile data.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Example 1
As shown in fig. 1, the embodiment provides a cross-sectional profile dimension reduction model training method, which is characterized in that,
step 101: constructing an initial section contour dimension reduction model; the section contour dimension reduction model is constructed based on a self-coding neural network;
step 102: and carrying out optimization training on the initial section contour dimension reduction model to obtain a trained section contour dimension reduction model.
In order to make the dimension reduction model of the profile provided by the embodiment more conform to the profile data of the strip material in the production process, before the dimension reduction model of the initial profile is constructed, the dimension reduction model of the profile of the strip material in the production process is collected, and noise is added to the profile data of the strip material to obtain a sample data set. And carrying out pretreatment of standardizing and eliminating repeated values on the sample data set to obtain a processed sample data set, and carrying out optimization training on the initial section contour dimension reduction model according to the processed sample data set.
Specifically, in the rolling process, the section profile data of the strip is collected and divided into three areas, wherein the left area is the convexity at a distance of 3mm, 5mm, 7mm, 10mm, 15mm, 20mm, 25mm, 30mm, 50mm, 75mm, 100mm, 200mm and 300mm from the left side edge of the strip, namely WS3, WS5, WS7, WS10, WS15, WS20, WS25, WS30, WS50, WS75, WS100, WS125, WS200 and WS300, and the total number is 14 discrete values; the right region is the convexity at 3mm, 5mm, 7mm, 10mm, 15mm, 20mm, 25mm, 30mm, 50mm, 75mm, 100mm, 200mm, 300mm from the right edge of the strip, i.e. DS3, DS5, DS7, DS10, DS15, DS20, DS25, DS30, DS50, DS75, DS100, DS125, DS200, DS300, for a total of 14 discrete values; the convexity at the center point is WSC and DSC, totaling 2 discrete values. As the rolling process proceeds, T sets of profile values with a dimension of 30 were collected in the rolling direction, expressed as:
because the value of central zone all can be detected when the left and right sides is detected, and two times of repeated numerical values appear in WSC and DSC promptly, consequently eliminate the repeated numerical value and guarantee sample data and the degree of closeness of actual production data, the data after the removal are:
because the production process of the strip is uninterrupted, the section profile data of each roll of the strip is stored in one file, and n files are selected as a sample data set, wherein R is { D ═ D 1 ′,D 2 ′,D 3 ′…,D n ' }, as shown in Table 1.
Table 1 data set R partial data (50 groups)
TABLE 1 continuation
TABLE 1 continuation
In order to make the network better fit the target and accelerate the convergence speed, the strip section profile data should be standardized: all profile values C in the data set R divided by the maximum value C in R max I.e. by(C is the original profile, C max Is the maximum value of the profile, C Sign board Normalized value) to [0, 1%]In the meantime. The normalized values for data set R are shown in table 2.
Table 2 data set R normalized partial data (50 groups)
TABLE 2 continuation
TABLE 2 continuation
Meanwhile, considering that the strip is in an unstable stage when rolling starts, the convexity meter cannot detect the profile of the section in certain channels, so that data is not completely lost, signals are interfered in the transmission process, so that the transmitted data is inaccurate, and finally the obtained data is noisy. In order to improve the accuracy of the cross-section profile dimension reduction model in practical application and improve the reconstruction capability of a self-encoder in the model, noise is added to the cross-section profile data of each roll of strip material in the embodiment. In order to fit a complex actual production working condition, gaussian noise is selected as noise, a noise reduction factor of the noise reduction self-encoder is set to be N, wherein N is 0.001 in the embodiment, and in subsequent model training, if a mean square error calculated by a trained model is large, the noise reduction factor can be adjusted.
After a sample data set which is subjected to the preprocessing of standardization and repeated value elimination and added with noise is obtained, the initial section contour dimension reduction model is subjected to optimization training by utilizing the sample data set.
First, as shown in fig. 5, constructing an initial profile dimension reduction model includes constructing an encoder and a decoder. Specifically, a noise reduction self-coding neural network is preset and comprises one to five layers which are connected in sequence, a coding part of the noise reduction self-coding device is formed from the first layer to the third layer, a decoding part is formed from the third layer to the fifth layer, and the connection mode of each layer of the self-coding device is full connection. The number of the neurons of the first layer inputlayer and the fifth layer outputlayer is consistent with the input variable of the strip section profile data, and the strip section profile data provided in the embodiment has C 1 -C 29 29 input variables are totally set, so that if the number of the neurons in the first layer and the fifth layer is set to be 29, the output variable C is set 1 '-C' 29 There are also 29. Setting the number of the neurons of the third layer encoder _3 as dimension l (1) of dimension reduction<l<28) The number of neurons in the second layer _2 and the fourth layer _4 is set as an initial value, and in this embodiment, the initial values of the second layer and the fourth layer are both set to 20.
Then, an activation function of each layer of the noise reduction self-coding neural network is selected, relu is selected from the activation functions of the first layer to the fifth layer, and tanh is selected as the activation function of the output layer. And finally, constructing an initial section profile dimension reduction model based on the noise reduction self-coding neural network.
And then training the initial section contour dimension reduction model: setting a reconstruction error function as a mean square error loss function mse, and then training the initial section contour dimension reduction model for preset iteration times to obtain a first model;
and judging whether the first model meets a preset precision threshold value or not according to the reconstruction error function, if so, taking the first model as a section contour dimension reduction model, otherwise, optimizing and adjusting parameters of the initial section contour dimension reduction model, and returning to the step of training the initial section contour dimension reduction model for preset iteration times.
Wherein, optimizing and adjusting the initial section contour dimension reduction model comprises:
(a) adjusting a network structure: adjusting the number of hidden layer layers of the initial section profile dimensionality reduction model and the number of neurons of each hidden layer;
(b) weight initialization: initializing the weight of each hidden layer by using a small random number;
(c) adjusting parameters: and adjusting batch training size batch-size, learning rate learning-rate and training step number Epoch of the initial cross-section contour dimension reduction model.
In order to better perform the training process, besides presetting the iteration number Epoch, the training parameters of the model also need to be set: determining the batch training size batch-size, setting the learning rate learning-rate, and selecting the optimizer. In this embodiment, the batch training size batch-size is specifically set to be 128, the learning rate learning-rate is 0.002, the iteration number Epoch of training is 150, and the optimizer chooses Adam.
Further, determining whether the first model meets a preset precision threshold according to the reconstruction error function includes:
taking partial data in the sample data set as training samples, and dividing the training samples into a training set and a verification set; and randomly extracting 90% of training samples as a training set train of the noise reduction self-coding neural network, wherein the training set train comprises 10000 groups, and the rest 10% of training samples are used as a verification set validity, and the total number of the verification sets is 1000 groups.
Training the initial section contour dimension reduction model by using the training set for preset iteration times to obtain a first model;
and judging whether the average error of the first model is smaller than a first preset threshold value according to the training error loss diagrams of the training set and the verification set, if so, the first model meets a preset precision threshold value, and otherwise, the first model does not meet the preset precision threshold value. As shown in fig. 2, the average error of the network is 0.002, in this embodiment, the first preset threshold is specifically set to be 0.01, and the first model satisfies the preset accuracy threshold.
In order to further improve the accuracy of model training and ensure the compression effect of the profile data, part of data in the sample data set except the training sample is used as a test set. And when the first model is smaller than a first preset threshold value, inputting the test set into the first model to obtain an output test set, and judging whether the mean square error of the test set and the output test set is smaller than a second preset threshold value, wherein if the mean square error of the test set and the output test set is smaller than the second preset threshold value, the first model meets a preset precision threshold value, and otherwise, the first model does not meet the preset precision threshold value.
In this embodiment, the input and output mean square error of the test set is calculated to be 8.33, the second preset threshold is 10, and the reconstruction accuracy of the model meets the preset condition and can reach a decimal point one bit. FIG. 3 is a comparison graph of the original value and the self-encoding reconstructed value of the strip section profile, and it can be seen from FIG. 3 that the position error with large curve fluctuation is large, but the compression precision is met. Fig. 4(a) (b) is a gray scale comparison graph of the reconstructed value and the original value of the section profile of a part of strip, and the degree of distortion of the noise reduction self-coding is very small by comparing the gray scales of a plurality of groups of section profile curves.
Therefore, after the initial cross-section profile dimension reduction model is trained, the trained cross-section profile dimension reduction model is obtained.
Example 2
As shown in fig. 6, the present embodiment provides a cross-sectional profile dimension reduction model training system, which includes:
the model building module M1 is used for building an initial section profile dimension reduction model;
an optimization training module M2, configured to perform optimization training on the initial profile dimensionality reduction model to obtain a trained profile dimensionality reduction model
Example 3
The embodiment provides a method for compressing strip section profile data, which uses the section profile dimension reduction model as described in embodiment 1, and the method includes:
and compressing the strip section profile data by using the section profile dimension reduction model.
Example 4
The embodiment provides a strip section profile data compression system, the system includes:
and the data compression module is used for compressing the strip material section profile data by using the section profile dimension reduction model.
The emphasis of each embodiment in the present specification is on the difference from the other embodiments, and the same and similar parts among the various embodiments may be referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (5)
1. A cross section contour dimension reduction model training method is characterized by comprising the following steps:
constructing an initial section contour dimension reduction model; the section contour dimension reduction model is constructed based on a self-coding neural network;
collecting strip section contour data in a production process, and adding noise into the strip section contour data to obtain a sample data set;
carrying out pretreatment of standardizing and eliminating repeated values on the sample data set to obtain a processed sample data set, and carrying out optimization training on the initial section contour dimension reduction model according to the processed sample data set;
carrying out optimization training on the initial section contour dimension reduction model to obtain a trained section contour dimension reduction model;
the optimization training of the initial section contour dimension reduction model comprises the following steps:
constructing a reconstruction error function;
training the initial section contour dimension reduction model for preset iteration times to obtain a first model;
judging whether the first model meets a preset precision threshold value or not according to the reconstruction error function, if so, taking the first model as a section contour dimension reduction model, otherwise, optimizing and adjusting parameters of the initial section contour dimension reduction model, and returning to the step of training the initial section contour dimension reduction model for preset iteration times;
the determining whether the first model meets a preset precision threshold according to the reconstruction error function includes:
taking partial data in the sample data set as training samples, and dividing the training samples into a training set and a verification set;
training the initial section contour dimension reduction model by using the training set for preset iteration times to obtain a first model;
judging whether the average error of the first model is smaller than a first preset threshold value according to the training error loss graphs of the training set and the verification set, if so, enabling the first model to meet a preset precision threshold value, otherwise, enabling the first model not to meet the preset precision threshold value;
the determining whether the first model meets a preset precision threshold according to the reconstruction error function further includes:
taking part of data in the sample data set except the training sample as a test set;
when the first model is smaller than a first preset threshold value, inputting the test set into the first model to obtain an output test set, and judging whether the mean square error of the test set and the output test set is smaller than a second preset threshold value, wherein if the mean square error of the test set and the output test set is smaller than the second preset threshold value, the first model meets a preset precision threshold value, otherwise, the first model does not meet the preset precision threshold value;
the optimizing and adjusting the parameters of the initial section profile dimension reduction model comprises the following steps:
adjusting the number of hidden layer layers of the initial section profile dimensionality reduction model and the number of neurons of each hidden layer;
initializing the weight of each hidden layer by using a small random number;
and adjusting the batch training size, the learning rate and the training steps of the initial section contour dimension reduction model.
2. The method for training the cross-sectional profile dimensionality reduction model according to claim 1, wherein the constructing the initial cross-sectional profile dimensionality reduction model comprises:
constructing a self-coding neural network comprising five layers, wherein the first layer to the fifth layer are connected in sequence in a full-connection mode;
setting the number of neurons of the first layer and the fifth layer according to the input variable of the strip section profile data;
setting the number of neurons in the third layer according to the dimensionality needing dimension reduction;
and setting the number of the neurons of the second layer and the fourth layer as an initial value.
3. The method for compressing the strip section profile data obtained by the training method of the section profile dimension reduction model according to claim 1, is characterized by comprising the following steps:
and compressing the strip section profile data by using the section profile dimension reduction model.
4. The system for compressing the strip section profile data obtained by the method for training the section profile dimension reduction model according to claim 1, is characterized by comprising the following steps:
and the data compression module is used for compressing the strip material section profile data by using the section profile dimension reduction model.
5. A system for training a cross-sectional profile dimension reduction model, the system comprising:
the model construction module is used for constructing an initial section contour dimension reduction model;
collecting strip section contour data in a production process, and adding noise into the strip section contour data to obtain a sample data set;
carrying out pretreatment of standardizing and eliminating repeated values on the sample data set to obtain a processed sample data set, and carrying out optimization training on the initial section contour dimension reduction model according to the processed sample data set;
the optimization training module is used for performing optimization training on the initial section contour dimension reduction model to obtain a trained section contour dimension reduction model;
the optimization training of the initial section contour dimension reduction model comprises the following steps:
constructing a reconstruction error function;
training the initial section contour dimension reduction model for preset iteration times to obtain a first model;
judging whether the first model meets a preset precision threshold value or not according to the reconstruction error function, if so, taking the first model as a section contour dimension reduction model, otherwise, optimizing and adjusting parameters of the initial section contour dimension reduction model, and returning to the step of training the initial section contour dimension reduction model for preset iteration times;
the determining whether the first model meets a preset precision threshold according to the reconstruction error function includes:
taking partial data in the sample data set as training samples, and dividing the training samples into a training set and a verification set;
training the initial section contour dimension reduction model by using the training set for preset iteration times to obtain a first model;
judging whether the average error of the first model is smaller than a first preset threshold value according to the training error loss graphs of the training set and the verification set, if so, enabling the first model to meet a preset precision threshold value, otherwise, enabling the first model not to meet the preset precision threshold value;
the determining whether the first model meets a preset precision threshold according to the reconstruction error function further includes:
taking part of data in the sample data set except the training sample as a test set;
when the first model is smaller than a first preset threshold value, inputting the test set into the first model to obtain an output test set, and judging whether the mean square error of the test set and the output test set is smaller than a second preset threshold value, wherein if the mean square error of the test set and the output test set is smaller than the second preset threshold value, the first model meets a preset precision threshold value, otherwise, the first model does not meet the preset precision threshold value;
the optimizing and adjusting the parameters of the initial section profile dimension reduction model comprises the following steps:
adjusting the number of hidden layer layers of the initial section profile dimensionality reduction model and the number of neurons of each hidden layer;
initializing the weight of each hidden layer by using a small random number;
and adjusting the batch training size, the learning rate and the training steps of the initial section contour dimension reduction model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110805792.XA CN113452379B (en) | 2021-07-16 | 2021-07-16 | Section contour dimension reduction model training method and system and data compression method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110805792.XA CN113452379B (en) | 2021-07-16 | 2021-07-16 | Section contour dimension reduction model training method and system and data compression method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113452379A CN113452379A (en) | 2021-09-28 |
CN113452379B true CN113452379B (en) | 2022-08-02 |
Family
ID=77816453
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110805792.XA Active CN113452379B (en) | 2021-07-16 | 2021-07-16 | Section contour dimension reduction model training method and system and data compression method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113452379B (en) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107038503A (en) * | 2017-04-18 | 2017-08-11 | 广东工业大学 | A kind of Demand Forecast method and system of shared equipment |
CN110110848A (en) * | 2019-05-05 | 2019-08-09 | 武汉烽火众智数字技术有限责任公司 | A kind of combination forecasting construction method and device |
CN111122989A (en) * | 2019-12-30 | 2020-05-08 | 厦门大学 | K-clustering intelligent selection microwave signal multipath interference suppression method |
CN111310920A (en) * | 2020-03-17 | 2020-06-19 | 无锡多纬智控科技有限公司 | Method for applying deep learning neural network technology to signal acquisition device |
CN112053333A (en) * | 2020-08-31 | 2020-12-08 | 中冶赛迪重庆信息技术有限公司 | Square billet detection method, system, equipment and medium based on machine vision |
CN112983608A (en) * | 2021-02-09 | 2021-06-18 | 联合汽车电子有限公司 | Particle trap carbon load calculation method and system, controller and storage medium |
Family Cites Families (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103886371A (en) * | 2014-03-28 | 2014-06-25 | 郑州大学 | Method for controlling component and thermal treatment technological process of pre-hardening plastic die steel |
CN105426962A (en) * | 2015-11-18 | 2016-03-23 | 长安大学 | Method for constructing and training dynamic neural network of incomplete recursive support |
CN105893637A (en) * | 2016-06-24 | 2016-08-24 | 四川大学 | Link prediction method in large-scale microblog heterogeneous information network |
CN110149333B (en) * | 2019-05-23 | 2021-06-29 | 桂林电子科技大学 | Network security situation assessment method based on SAE + BPNN |
US20200387789A1 (en) * | 2019-06-06 | 2020-12-10 | Riskfuel Analytics Inc. | Neural network training |
US11615301B2 (en) * | 2019-09-03 | 2023-03-28 | Microsoft Technology Licensing, Llc | Lossless exponent and lossy mantissa weight compression for training deep neural networks |
CN111461224B (en) * | 2020-04-01 | 2022-08-16 | 西安交通大学 | Phase data unwrapping method based on residual self-coding neural network |
CN112115024B (en) * | 2020-09-03 | 2023-07-18 | 上海上讯信息技术股份有限公司 | Training method and device for fault prediction neural network model |
CN113113030B (en) * | 2021-03-22 | 2022-03-22 | 浙江大学 | High-dimensional damaged data wireless transmission method based on noise reduction self-encoder |
-
2021
- 2021-07-16 CN CN202110805792.XA patent/CN113452379B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107038503A (en) * | 2017-04-18 | 2017-08-11 | 广东工业大学 | A kind of Demand Forecast method and system of shared equipment |
CN110110848A (en) * | 2019-05-05 | 2019-08-09 | 武汉烽火众智数字技术有限责任公司 | A kind of combination forecasting construction method and device |
CN111122989A (en) * | 2019-12-30 | 2020-05-08 | 厦门大学 | K-clustering intelligent selection microwave signal multipath interference suppression method |
CN111310920A (en) * | 2020-03-17 | 2020-06-19 | 无锡多纬智控科技有限公司 | Method for applying deep learning neural network technology to signal acquisition device |
CN112053333A (en) * | 2020-08-31 | 2020-12-08 | 中冶赛迪重庆信息技术有限公司 | Square billet detection method, system, equipment and medium based on machine vision |
CN112983608A (en) * | 2021-02-09 | 2021-06-18 | 联合汽车电子有限公司 | Particle trap carbon load calculation method and system, controller and storage medium |
Non-Patent Citations (2)
Title |
---|
Coiflets, artificial neural networks and predictive coding based hybrid image compression methodology;Siripurapu Sridhar等;《IEEE》;20141016;全文 * |
基于BP神经网络的物联网上市企业经营绩效研究;江旭;《中国优秀硕士学位论文全文数据库 (信息科技辑)》;20200115;全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN113452379A (en) | 2021-09-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107590565B (en) | Method and device for constructing building energy consumption prediction model | |
CN109858408B (en) | Ultrasonic signal processing method based on self-encoder | |
CN117434153B (en) | Road nondestructive testing method and system based on ultrasonic technology | |
CN115659160A (en) | Data quality measurement method for optimizing digital twin model | |
CN110569566A (en) | Method for predicting mechanical property of plate strip | |
CN112381790A (en) | Abnormal image detection method based on depth self-coding | |
CN104070083A (en) | Method for measuring rotating speed of guiding disc of perforating machine based on integrated PCA-ELM (Principal Component Analysis)-(Extrem Learning Machine) method | |
CN110879927A (en) | Sea clutter amplitude statistical distribution field modeling method for sea target detection | |
CN103203670B (en) | Grinding chatter prediction method based on maximum information entropy and directional divergence | |
CN113452379B (en) | Section contour dimension reduction model training method and system and data compression method and system | |
CN115018790A (en) | Workpiece surface defect detection method based on anomaly detection | |
CN114239810A (en) | Milling cutter wear prediction method based on improved PCANet model | |
CN110751400B (en) | Risk assessment method and device | |
CN110880192B (en) | Image DCT coefficient distribution fitting method based on probability density function dictionary | |
CN113609766A (en) | Soft measurement method based on depth probability latent model | |
CN111104640A (en) | Rainfall observation and evaluation method and system based on analytic hierarchy process | |
WO2022188574A1 (en) | Deep learning method and apparatus for regression task | |
CN115099135A (en) | Improved artificial neural network multi-type operation power consumption prediction method | |
CN111858699A (en) | Time series correlation detection method, equipment and storage medium | |
CN115270895B (en) | Fault detection method for diesel engine | |
CN116630814B (en) | Quick positioning and evaluating method for building disasters based on machine learning | |
Li et al. | Image edge detection algorithm based on GM (1, 1, C) | |
CN109521176B (en) | Virtual water quality monitoring method based on improved deep extreme learning machine | |
CN117808810B (en) | Hobbing cutter abrasion image recognition and measurement method and system | |
CN109241041B (en) | Preprocessing method and device for big data of injection molding equipment |
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 | ||
TR01 | Transfer of patent right |
Effective date of registration: 20240207 Address after: Unit 901-01, 8th Floor, Unit 1, Building 1, No. 3 Changtong Road, Chaoyang District, Beijing, 100000 Patentee after: Beijing Mitian Technology Co.,Ltd. Country or region after: China Address before: No.438, west section of Hebei Street, Haigang District, Qinhuangdao City, Hebei Province Patentee before: Yanshan University Country or region before: China |
|
TR01 | Transfer of patent right |