CN111046080A - Carbon fiber precursor monitoring data preprocessing method based on convolution denoising autoencoder - Google Patents
Carbon fiber precursor monitoring data preprocessing method based on convolution denoising autoencoder Download PDFInfo
- Publication number
- CN111046080A CN111046080A CN201911238154.3A CN201911238154A CN111046080A CN 111046080 A CN111046080 A CN 111046080A CN 201911238154 A CN201911238154 A CN 201911238154A CN 111046080 A CN111046080 A CN 111046080A
- Authority
- CN
- China
- Prior art keywords
- data
- data set
- layer
- carbon fiber
- layers
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000012544 monitoring process Methods 0.000 title claims abstract description 64
- 229920000049 Carbon (fiber) Polymers 0.000 title claims abstract description 59
- 239000004917 carbon fiber Substances 0.000 title claims abstract description 59
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 title claims abstract description 58
- 238000000034 method Methods 0.000 title claims abstract description 58
- 239000002243 precursor Substances 0.000 title claims abstract description 45
- 238000007781 pre-processing Methods 0.000 title claims abstract description 33
- 238000012549 training Methods 0.000 claims abstract description 23
- 238000012217 deletion Methods 0.000 claims abstract description 22
- 230000037430 deletion Effects 0.000 claims abstract description 22
- 238000004519 manufacturing process Methods 0.000 claims abstract description 20
- 230000008569 process Effects 0.000 claims abstract description 20
- 230000006870 function Effects 0.000 claims description 35
- 238000011049 filling Methods 0.000 claims description 29
- 238000010606 normalization Methods 0.000 claims description 16
- 230000004913 activation Effects 0.000 claims description 12
- 210000002569 neuron Anatomy 0.000 claims description 9
- 238000011176 pooling Methods 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 9
- 239000000654 additive Substances 0.000 claims description 6
- 230000000996 additive effect Effects 0.000 claims description 6
- 239000013598 vector Substances 0.000 claims description 6
- 238000011068 loading method Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000002203 pretreatment Methods 0.000 claims 1
- 238000007405 data analysis Methods 0.000 description 6
- 238000007380 fibre production Methods 0.000 description 6
- 230000008901 benefit Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 238000009987 spinning Methods 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000001035 drying Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000000835 fiber Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 239000000243 solution Substances 0.000 description 2
- 238000007619 statistical method Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 238000005406 washing Methods 0.000 description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 2
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 229910000831 Steel Inorganic materials 0.000 description 1
- XAGFODPZIPBFFR-UHFFFAOYSA-N aluminium Chemical compound [Al] XAGFODPZIPBFFR-UHFFFAOYSA-N 0.000 description 1
- 229910052782 aluminium Inorganic materials 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 229910052799 carbon Inorganic materials 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 230000007123 defense Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 239000002657 fibrous material Substances 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000008676 import Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012806 monitoring device Methods 0.000 description 1
- 238000004806 packaging method and process Methods 0.000 description 1
- 229920002239 polyacrylonitrile Polymers 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000010025 steaming Methods 0.000 description 1
- 239000010959 steel Substances 0.000 description 1
- 239000011550 stock solution Substances 0.000 description 1
- 239000004753 textile Substances 0.000 description 1
- 238000004804 winding Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2465—Query processing support for facilitating data mining operations in structured databases
-
- 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
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Fuzzy Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Treatment Of Fiber Materials (AREA)
Abstract
The invention discloses a carbon fiber precursor monitoring data preprocessing method based on a convolution denoising autoencoder, which comprises the following steps: 1) collecting monitoring data of carbon fiber precursor production, and arranging the data into a data set according to a time sequence and a process sequence; 2) selecting a sample without deletion in the data set as a target data set; carrying out random missing operation on the target data set to obtain a training data set; 3) establishing a convolution denoising autoencoder model, and training the model by using the training data set and the target data set stored in the step 2); 4) preprocessing the precursor monitoring data to be processed by utilizing the convolutional denoising self-encoder model trained in the step 3). The invention solves the problems of deletion and outlier in the carbon fiber monitoring data by designing the carbon fiber precursor monitoring data preprocessing method based on the convolution denoising autoencoder, and can quickly finish the preprocessing of the carbon fiber monitoring data.
Description
Technical Field
The invention relates to a carbon fiber production technology, in particular to a carbon fiber precursor monitoring data preprocessing method based on a convolution denoising autoencoder.
Background
The carbon fiber is a novel high-strength and high-modulus fiber material with the carbon content of more than 90 percent, is lighter than aluminum in mass, has higher strength than steel, and has the characteristics of corrosion resistance and fatigue resistance. Meanwhile, the fiber has the soft processability of textile fiber, and is an important material in the aspects of national defense, military industry and civil use. The polyacrylonitrile-based carbon fiber is the carbon fiber with the highest market share at present due to the simple process, low cost and excellent performance. The research and production of carbon fiber in China are slow to start, and the current domestic high-performance carbon fiber products are limited by process technology and cannot realize comprehensive industrial production, so that more than 80 percent of high-performance carbon fiber in China depends on import.
The quality of carbon fiber precursors is a bottleneck for restricting the improvement of carbon fibers. In order to improve the quality of the precursor and ensure the production stability, the real-time monitoring of the production process of the carbon fiber precursor is necessary. In recent years, the monitoring problem of carbon fiber production process is more and more regarded by the industry. With the development of intelligent manufacturing and big data, more and more monitoring devices and sensors are deployed in the carbon fiber production process, so that a large amount of production process monitoring data is obtained. However, due to network transmission problems, sensor faults and other human factors, data quality problems such as noise, missing and the like often exist in carbon fiber monitoring data, and the characteristics of large data volume and high dimensionality bring great inconvenience to subsequent data analysis or modeling problems. Data loss refers to a phenomenon in which a part of a time period is lost in time-series data or a part of a dimension is lost. Data noise refers to the presence of erroneous values in the data or significant deviations from expected outliers. In carbon fiber monitoring data, prominent data quality issues are lack of dimensional data and insignificant outliers.
In the traditional data preprocessing method, data are preprocessed by a statistical knowledge or machine learning method. For the processing of missing data, although the filling speed is high by simple statistical methods (such as mean filling and median filling), the filling result deviates from the distribution of the original data to a certain extent; complex filling methods (such as regression filling, KNNI and the like) have low filling efficiency on large-scale data sets; for noise existing in monitoring data, a smoothing filtering method adopted at present smoothes data flow and removes abrupt change data. With the development of deep learning and the increasing amount of monitoring data, more and more deep learning techniques are applied to the field of data preprocessing. Compared with the traditional method, the deep learning has remarkable advantages in processing the problems of high dimensionality and large size.
The production environment of the carbon fiber precursor is complex, the processes are multiple, a complete precursor spinning process comprises the steps of preparation of stock solution, spinning, washing, drafting, oiling, drying and the like, and meanwhile, strong coupling relations exist among technological parameters, so that the monitoring data of the carbon fiber precursor production has the characteristics of high dimensionality, large volume, high value, timeliness, coupling and miscellaneous property. On one hand, the monitoring amount at a certain moment is the continuation of the change condition of the monitoring amount at the last moment; on the other hand, the monitoring amount is influenced by other monitoring amounts at the same time. The complex correlation brings certain difficulties to data preprocessing and subsequent data mining work. However, the existing data preprocessing method, especially the solution to the lack of carbon fiber monitoring and noise interference, lacks attention to the characteristics of the carbon fiber monitoring data, and the common statistical method has low processing efficiency and poor effect.
In order to solve the problems of deletion and noise contained in data quality in carbon fiber production monitoring data and the problems caused by the characteristics of large data volume and high dimensionality, the invention designs a carbon fiber precursor monitoring data preprocessing method based on a convolution denoising autoencoder. The method solves the problems of deletion and noise in the carbon fiber monitoring data through the convolution denoising self-encoder model, and has great advantages on the carbon fiber monitoring data with high dimension and large volume.
Disclosure of Invention
The invention aims to solve the technical problem of providing a carbon fiber precursor monitoring data preprocessing method based on a convolution denoising autoencoder aiming at the defects in the prior art.
The technical scheme adopted by the invention for solving the technical problems is as follows: a carbon fiber precursor monitoring data preprocessing method based on a convolution denoising autoencoder comprises the following steps:
1) collecting monitoring data of carbon fiber precursor production, and arranging the data into a data set according to a time sequence and a process sequence;
2) selecting a sample without deletion in the data set as a target data set; carrying out random missing operation on the target data set to obtain a training data set;
3) establishing a convolution denoising autoencoder model, and training the model by using the training data set and the target data set stored in the step 2);
the convolution denoising self-encoder model is as follows:
the convolution denoising self-encoder comprises 10 layers including a full connection layer, a 2D convolution layer, a maximum pooling layer, an upper sampling layer and a deconvolution layer,
wherein, the 1 st layer is a full-connection layer, the 2 nd and 4 th layers are 2D convolutional layers, the 3 rd and 5 th layers are maximum pooling layers, the 6 th and 8 th layers are up-sampling layers, the 7 th and 9 th layers are deconvolution layers, the 10 th layer is an output layer, and the output layer is a full-connection layer;
4) preprocessing the precursor monitoring data to be processed by utilizing the convolutional denoising self-encoder model trained in the step 3).
According to the scheme, the data in the step 1) are arranged into a data set according to a time sequence and a process sequence, and the method specifically comprises the following steps:
1.1) collecting real-time monitoring data of the production process of carbon fiber precursors, generating a data set X ═ X according to the data types of temperature, pressure and flow by taking time as a collection sequenceiWherein the operation data value of the system at each moment is used as one row of the matrix, i is more than or equal to 1 and less than or equal to n, n is the number of data variables, and the monitoring data comprises each of the raw yarn productionThe time period, for each data type, the data set is generated as:
X＝[X1，X2，X3，…，Xi…，Xn]
1.2) reordering the features of the data set according to the following principle:
arranging the characteristics of the data set according to the process sequence of carbon fiber precursor production;
the monitoring data of the same type in the same procedure are adjacent;
1.3) uploading the sorted data set to a cloud platform.
According to the scheme, the specific steps of obtaining the training data set in the step 2) are as follows:
2.1) selecting a data subset X without deletion in the data set sorted in the step 1);
2.2) carrying out normalization processing on the data set X to serve as a target data set D _ output, wherein the normalization method is as follows:
wherein X is the data in the selected non-missing data set X, and X' is the data of the normalized data set X;
2.3) carrying out random deletion operation on the data set by using a random zero setting method, wherein the deletion proportion is set according to the deletion condition of the carbon fiber precursor monitoring data as required, so as to obtain a damaged data set D _ corruption;
and 2.4) normalizing the damaged data set D _ corruption to obtain an input data set D _ input.
According to the above scheme, before the random null-steering operation is performed on the data set by using the random null-steering method in the step 2.3), the method further comprises an operation of adding gaussian noise to the data set:
adding additive white Gaussian noise to X;
X′i＝Xi+μ*Zi～N(0，1)
where μ is a noise factor used to control the amplitude of the additive noise.
According to the scheme, in the step 3), the loss function adopted in the model training is selected as the mean square error of the reconstructed sample D _ train and the target sample D _ output;
wherein x is the target sample,
to reconstruct the sample.
According to the scheme, in the step 4), preprocessing is performed on the precursor monitoring data to be processed, and the preprocessing comprises the following specific steps:
4.1) sorting the data into a data set by taking days as a unit, rearranging the data set, and replacing missing values with zero values;
4.2) normalizing the data set;
4.3) loading the convolution denoising self-encoder model in the step 3), taking the data after normalization processing as input, and obtaining reconstructed data by using the model;
and 4.4) performing inverse normalization on the reconstructed data to obtain the preprocessed precursor monitoring data.
According to the scheme, the convolution denoising self-encoder has the following specific structure:
3.1) the layer 1 is a fully-connected layer, the number of neurons is 256, the activation function is a relu function, and then a feature vector output by the first layer is constructed into a feature map by utilizing a Reshape function, wherein the size is 16 × 1;
ReLu(x)＝max(0，x)
h1＝ReLu(w1h0+b1)
wherein h is0Is an input layer, w1Is the total connection layer neuron weight, b1Is an offset.
3.2) the 2 nd and 4 th layers are 2D convolutional layers, the depths of the convolutional layers are respectively 16 and 8, the sizes of convolution kernels are 3 x 3, the activation functions are all relu functions, the filling mode is SAME filling, and the purpose of filling is to keep the sizes of the characteristic graphs unchanged;
h2，a＝ReLu(h1*w2，a+b2，a)
wherein h is2，a，w2，aIs the a-th convolution area and convolution kernel of the convolution layer;
3.3) the 3 rd and 5 th layers are maximum pooling layers, the size of the pools is 2 x 2, and the filling is SAME filling;
h3，a＝down(h2，a)
wherein h is3，aA down-sampling result of the corresponding volume area;
3.4) the 6 th and 8 th layers are upper sampling layers with the size of 2 x 2;
3.5) the 7 th layer and the 9 th layer are deconvolution layers, the number of convolution kernels in the deconvolution layers is 16 and 1 respectively, the size of the convolution kernels is 3 x 3, the activation functions are all relu functions, and the filling is SAME filling;
3.6) the 10 th layer is an output layer, the output layer is a full-connection layer, the number of the neurons is 195, the activation function is a relu function, and before entering the full-connection layer, the feature map output by the 9 th layer needs to be expanded to be reconstructed into a feature vector.
The invention has the following beneficial effects:
1. in the existing data preprocessing process, the complex correlation existing in the carbon fiber monitoring data is ignored, so that the preprocessing effect is poor. Aiming at the problems, the convolutional layer and the pooling layer are added into the denoising self-encoder model, the data is better reconstructed by utilizing the correlation relationship among the data, and great advantages are shown in the preprocessing effect.
2. In the existing data preprocessing process, denoising and deletion filling are divided into two independent steps, and due to the problems of high dimension and large volume of monitoring data, the traditional method has the defects of low efficiency or low precision. Aiming at the problems, the invention designs the carbon fiber precursor monitoring data preprocessing method based on the convolution denoising autoencoder, so that denoising and deletion filling are effectively fused, the problems of deletion and outlier existing in the carbon fiber monitoring data are solved, and the preprocessing of the carbon fiber monitoring data can be quickly completed.
Drawings
The invention will be further described with reference to the accompanying drawings and examples, in which:
FIG. 1 is a flow chart of a method of an embodiment of the present invention;
FIG. 2 is a network structure diagram of a convolutional denoising autoencoder according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a denoised self-encoder according to an embodiment of the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
As shown in fig. 1, a method for preprocessing carbon fiber precursor monitoring data based on a convolution denoising autoencoder includes the following steps:
step 1: the method comprises the steps of collecting carbon fiber precursor production monitoring data from a real-time database of a carbon fiber production plant, uploading the carbon fiber precursor production monitoring data to a big data analysis platform, and sorting the data into a data set according to a time sequence and a process sequence. The method specifically comprises the following steps:
1.1: collecting real-time monitoring data of the production process of carbon fiber precursors from a real-time database of a carbon fiber production plant, generating a data set X (X) aiming at data types such as temperature, pressure, flow and the like by taking time as an acquisition sequenceiThe operation data value of the system at each moment is used as one row of the matrix, wherein i is more than or equal to 1 and less than or equal to n, n is the number of data variables, monitoring data needs to contain each time period of protofilament production as much as possible, and generation of a data set aiming at various data types can be expressed as follows:
X＝[X1，X2，X3，…，Xi…，Xn]
1.2: the features of the data set are reordered as follows. The principle is followed as follows:
① the characteristics of the data set are arranged according to the process sequence of carbon fiber precursor production (the process sequence is spinning, water washing, water drawing, oiling, drying, steaming and drawing, and winding).
② the same type of monitored data within the same process are adjacent.
1.3: and uploading the sorted data set to a cloud platform.
Step 2: selecting a sample without loss in a data set on a big data platform as a target data set; randomly damaging a target data set to be used as a training data set, and specifically comprising the following steps of:
2.1 selecting data X without deletion in the data set sorted in the step 1;
2.2, carrying out normalization processing on the data set X to serve as a target data set D _ output, and storing the normalization model on a big data analysis platform. The normalization method is as follows:
wherein X is the selected defect-free data set and X' is the normalized data
2.3 adding additive white Gaussian noise to X;
X′i＝Xi+μ*Zi～N(0，1)
where μ is the noise factor, which controls the amplitude of the additive noise.
2.4, performing random deletion operation on the data set added with Gaussian noise by using a random zero setting method, wherein the deletion proportion is set to be 0.1 according to the deletion condition of the monitoring data of the carbon fiber precursors, so as to obtain a damaged data set D _ corruption;
and 2.5, normalizing the damaged data set D _ corruption by using the stored normalization model to obtain an input data set D _ input.
And step 3: designing a convolution denoising self-encoder model, training the model by using the training data set and the target data set stored in the step 2, selecting the model with the best effect, storing the model as an HDF5 file as a final model, and deploying the model to a big data analysis platform.
The method specifically comprises the following steps:
3.1 design convolution denoising autoencoder, as shown in fig. 2, the autoencoder has 10 layers in total, and is composed of a full connection layer, a 2D convolution layer, a maximum pooling layer, an up-sampling layer and an anti-convolution layer, and the specific structure is as follows:
3.1.1 layer 1 is the fully-connected layer, the number of neurons is 256, and the activation function is the relu function. Compared with the sigmoid function and the tanh function, the relu function has low computational complexity and solves the problem of gradient disappearance. The feature vectors output by the first layer are then constructed into a feature map using the Reshape function, with a size of 16 x 1, which is done to accommodate the operation of subsequent convolutional layers.
ReLu(x)＝max(0，x)
h1＝ReLu(w1h0+b1)
Wherein h is0Is an input layer, w1Is the total connection layer neuron weight, b1Is an offset.
3.1.2 layers 2 and 4 are 2D convolutional layers, the depths of the convolutional layers are respectively 16 and 8, the sizes of convolution kernels are 3 x 3, and the activation functions are all relu functions. The filling mode is SAME filling. The purpose of the padding is to keep the size of the feature map constant.
h2，a＝ReLu(h1*w2，a+b2，a)
Wherein h is2，a，w2，aIs the a-th convolution area of the convolution layer and the convolution kernel.
3.1.3 layers 3 and 5 are the largest pooling layers, the pool size is 2 x 2, and the filling is SAME filling.
h3，a＝down(h2，a)
Wherein h is3，aIs the down-sampled result of the corresponding area of volume.
3.1.4 layers 6 and 8 are upsampled layers with a size of 2 x 2.
3.1.5 layers 7 and 9 are deconvolution layers, the number of convolution kernels in the deconvolution layers is 16 and 1 respectively, the size of the convolution kernels is 3 x 3, the activation functions are all relu functions, and the filling is SAME filling.
3.1.6 layer 10 is the output layer, the output layer is the full connection layer, its neuron number is 195, the activation function is relu function. Before entering the fully connected layer, the feature map output by the layer 9 needs to be expanded and reconstructed into a feature vector.
3.2 training the convolutional autocoder.
3.2.1 the input dataset is D _ input and the target dataset is D _ output.
3.2.2 in the training process, randomly selecting 20% of data in the training set as a verification set, wherein the iteration number is 200; and when the loss function of the training set tends to be stable and does not significantly decrease along with the iteration number any more, and the loss function of the verification set begins to show a rising trend, the training is proved to be finished.
3.2.3 the optimization algorithm is selected as Adam algorithm, Adam is a first-order optimization algorithm capable of replacing the traditional random gradient descent process, the neural network weight can be updated iteratively based on training data, and the initial learning rate is 0.01;
3.2.4 the loss function is chosen as the mean square error of the reconstructed sample D _ train and the target sample D _ output.
Wherein x is the target sample,
to reconstruct the sample.
3.3 storing the trained model and weights in an HDF5 file, wherein the file contains information such as the structure, weights, training configuration and the like of the model, and packaging and storing all the information on the cloud platform.
And 4, step 4: utilizing the preprocessing model trained in the step 3 to preprocess subsequent precursor monitoring data, storing the processed data into a data warehouse of a big data analysis platform to achieve the purpose of preprocessing, and specifically comprising the following steps:
4.1, sorting the data into a data set by taking days as a unit, rearranging the data set, and replacing missing values with zero values;
4.2, normalizing the data set by using a normalization model stored by a big data analysis platform;
4.3, loading the convolution denoising self-encoder model stored in the step 3, taking the data after normalization processing as input, and obtaining reconstructed data by using the model;
and 4.4, performing inverse normalization on the reconstructed data by using the normalization model to obtain preprocessed data, and storing the preprocessed data in a data warehouse.
It will be understood that modifications and variations can be made by persons skilled in the art in light of the above teachings and all such modifications and variations are intended to be included within the scope of the invention as defined in the appended claims.
Claims (7)
1. A carbon fiber precursor monitoring data preprocessing method based on a convolution denoising autoencoder is characterized by comprising the following steps:
1) collecting monitoring data of carbon fiber precursor production, and arranging the data into a data set according to a time sequence and a process sequence;
2) selecting a sample without deletion in the data set as a target data set; carrying out random missing operation on the target data set to obtain a training data set;
3) establishing a convolution denoising autoencoder model, and training the model by using the training data set and the target data set stored in the step 2);
the convolution denoising self-encoder model is as follows:
the convolution denoising self-encoder comprises 10 layers including a full connection layer, a 2D convolution layer, a maximum pooling layer, an upper sampling layer and a deconvolution layer,
wherein, the 1 st layer is a full-connection layer, the 2 nd and 4 th layers are 2D convolutional layers, the 3 rd and 5 th layers are maximum pooling layers, the 6 th and 8 th layers are up-sampling layers, the 7 th and 9 th layers are deconvolution layers, the 10 th layer is an output layer, and the output layer is a full-connection layer;
4) preprocessing the precursor monitoring data to be processed by utilizing the convolutional denoising self-encoder model trained in the step 3).
2. The pretreatment method of carbon fiber precursor monitoring data according to claim 1, wherein the data in step 1) are sorted into data sets according to a time sequence and a process sequence, and the data sets are as follows:
1.1) collecting real-time monitoring data of the production process of carbon fiber precursors, generating a data set X ═ X according to the data types of temperature, pressure and flow by taking time as a collection sequenceiAnd the operation data value of the system at each moment is taken as one row of the matrix, wherein i is more than or equal to 1 and less than or equal to n, n is the number of data variables, the monitoring data comprises each time period of protofilament production, and the generation of data sets aiming at various data types is represented as follows:
X＝[X1，X2，X3，…，Xi…，Xn]
1.2) reordering the features of the data set according to the following principle:
arranging the characteristics of the data set according to the process sequence of carbon fiber precursor production;
the monitoring data of the same type in the same procedure are adjacent;
1.3) uploading the sorted data set to a cloud platform.
3. The method for preprocessing carbon fiber precursor monitoring data according to claim 1, wherein the specific steps of obtaining the training data set in step 2) are as follows:
2.1) selecting a data subset X without deletion in the data set sorted in the step 1);
2.2) carrying out normalization processing on the data set X to serve as a target data set D _ output, wherein the normalization method is as follows:
wherein X is the data in the selected non-missing data set X, and X' is the data of the normalized data set X;
2.3) carrying out random deletion operation on the data set by using a random zero setting method, wherein the deletion proportion is set according to the deletion condition of the carbon fiber precursor monitoring data as required, so as to obtain a damaged data set D _ corruption;
and 2.4) normalizing the damaged data set D _ corruption to obtain an input data set D _ input.
4. The method for preprocessing carbon fiber precursor monitoring data according to claim 3, wherein before the random deletion operation is performed on the data set by using a random zero setting method in the step 2.3), the method further comprises an operation of adding Gaussian noise to the data set:
adding additive white Gaussian noise to X;
X′i＝Xi+μ*Zi～N(0，1)
where μ is a noise factor used to control the amplitude of the additive noise.
5. The method for preprocessing carbon fiber precursor monitoring data according to claim 1, wherein in the step 3), the loss function adopted in the model training is selected as the mean square error of the reconstructed sample D _ train and the target sample D _ output;
wherein x is the target sample,
to reconstruct the sample.
6. The method for preprocessing carbon fiber precursor monitoring data according to claim 1, wherein in the step 4), the precursor monitoring data to be processed is preprocessed, specifically as follows:
4.1) sorting the data into a data set by taking days as a unit, rearranging the data set, and replacing missing values with zero values;
4.2) normalizing the data set;
4.3) loading the convolution denoising self-encoder model in the step 3), taking the data after normalization processing as input, and obtaining reconstructed data by using the model;
and 4.4) performing inverse normalization on the reconstructed data to obtain the preprocessed precursor monitoring data.
7. The method for preprocessing carbon fiber precursor monitoring data as claimed in claim 1, wherein in the step 3), the convolutional denoising autoencoder has the following specific structure:
3.1) the layer 1 is a fully-connected layer, the number of neurons is 256, the activation function is a relu function, and then a feature vector output by the first layer is constructed into a feature map by utilizing a Reshape function, wherein the size is 16 × 1;
ReLu(x)＝max(0，x)
h1＝ReLu(w1h0+b1)
wherein h is0Is an input layer, w1Is the total connection layer neuron weight, b1Is an offset.
3.2) the 2 nd and 4 th layers are 2D convolutional layers, the depths of the convolutional layers are respectively 16 and 8, the sizes of convolution kernels are 3 x 3, the activation functions are all relu functions, the filling mode is SAME filling, and the purpose of filling is to keep the sizes of the characteristic graphs unchanged;
h2，a＝ReLu(h1*w2，a+b2，a)
wherein h is2，a，w2，aIs the a-th convolution area and convolution kernel of the convolution layer;
3.3) the 3 rd and 5 th layers are maximum pooling layers, the size of the pools is 2 x 2, and the filling is SAME filling;
h3，a＝down(h2，a)
wherein h is3，aA down-sampling result of the corresponding volume area;
3.4) the 6 th and 8 th layers are upper sampling layers with the size of 2 x 2;
3.5) the 7 th layer and the 9 th layer are deconvolution layers, the number of convolution kernels in the deconvolution layers is 16 and 1 respectively, the size of the convolution kernels is 3 x 3, the activation functions are all relu functions, and the filling is SAME filling;
3.6) the 10 th layer is an output layer, the output layer is a full-connection layer, the number of the neurons is 195, the activation function is a relu function, and before entering the full-connection layer, the feature map output by the 9 th layer needs to be expanded to be reconstructed into a feature vector.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911238154.3A CN111046080A (en) | 2019-12-05 | 2019-12-05 | Carbon fiber precursor monitoring data preprocessing method based on convolution denoising autoencoder |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911238154.3A CN111046080A (en) | 2019-12-05 | 2019-12-05 | Carbon fiber precursor monitoring data preprocessing method based on convolution denoising autoencoder |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111046080A true CN111046080A (en) | 2020-04-21 |
Family
ID=70234782
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911238154.3A Pending CN111046080A (en) | 2019-12-05 | 2019-12-05 | Carbon fiber precursor monitoring data preprocessing method based on convolution denoising autoencoder |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111046080A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114109456A (en) * | 2020-08-28 | 2022-03-01 | 神华神东煤炭集团有限责任公司 | Monitoring and early warning method and system for coal mine underground reservoir coal pillar dam body structure |
| CN114417923A (en) * | 2022-01-13 | 2022-04-29 | 哈尔滨工业大学 | Vehicle induced stress robust correlation mapping method based on convolution noise reduction self-encoder |
| CN116092614A (en) * | 2023-03-06 | 2023-05-09 | 山东大学 | Carbon fiber precursor preparation simulation method based on hybrid neural network |
| CN117116476A (en) * | 2023-07-04 | 2023-11-24 | 中国医学科学院阜外医院 | Downstream task prediction method and device and computer readable storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109213753A (en) * | 2018-08-14 | 2019-01-15 | 西安理工大学 | A kind of industrial system monitoring data restoration methods based on online PCA |
| CN109815223A (en) * | 2019-01-21 | 2019-05-28 | 北京科技大学 | A kind of complementing method and complementing device for industry monitoring shortage of data |
| CN109829903A (en) * | 2019-01-28 | 2019-05-31 | 合肥工业大学 | A kind of chip surface defect inspection method based on convolution denoising self-encoding encoder |
| CN109887290A (en) * | 2019-03-30 | 2019-06-14 | 西安电子科技大学 | Traffic flow forecasting method based on equilibrium index exponential smoothing and stack self-encoding encoder |
| CN109978079A (en) * | 2019-04-10 | 2019-07-05 | 东北电力大学 | A kind of data cleaning method of improved storehouse noise reduction self-encoding encoder |
-
2019
- 2019-12-05 CN CN201911238154.3A patent/CN111046080A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109213753A (en) * | 2018-08-14 | 2019-01-15 | 西安理工大学 | A kind of industrial system monitoring data restoration methods based on online PCA |
| CN109815223A (en) * | 2019-01-21 | 2019-05-28 | 北京科技大学 | A kind of complementing method and complementing device for industry monitoring shortage of data |
| CN109829903A (en) * | 2019-01-28 | 2019-05-31 | 合肥工业大学 | A kind of chip surface defect inspection method based on convolution denoising self-encoding encoder |
| CN109887290A (en) * | 2019-03-30 | 2019-06-14 | 西安电子科技大学 | Traffic flow forecasting method based on equilibrium index exponential smoothing and stack self-encoding encoder |
| CN109978079A (en) * | 2019-04-10 | 2019-07-05 | 东北电力大学 | A kind of data cleaning method of improved storehouse noise reduction self-encoding encoder |
Non-Patent Citations (10)
| Title |
|---|
| BO LI等: "DENOISING CONVOLUTIONAL AUTOENCODER BASED B-MODE ULTRASOUND TONGUE IMAGE FEATURE EXTRACTION" * |
| JANEBHOP SAWAENGCHOB等: "A Fast Convolutional Denoising Autoencoder Based Extreme Learning Machine" * |
| JUNJIE CHEN等: "Sparse Convolutional Denoising Autoencoders for Genotype Imputation" * |
| LOVEDEEP GONDARA等: "Medical image denoising using convolutional denoising autoencoders" * |
| LOVEDEEP GONDARA等: "Multiple Imputation using Denoising Autoencoders" * |
| PASCAL VINCENT等: "Extracting and Composing Robust Features with Denoising Autoencoders" * |
| PASCAL VINCENT等: "Pascal Vincent等 Extracting and Composing Robust Features with Denoising Autoencoders" * |
| 代杰杰等: "基于栈式降噪自编码器的输变电设备状态数据清洗方法" * |
| 殷敬伟等: "基于降噪自编码器的水声信号增强研究" * |
| 熊维: "面向视觉特征表达的深度学习算法研究" * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114109456A (en) * | 2020-08-28 | 2022-03-01 | 神华神东煤炭集团有限责任公司 | Monitoring and early warning method and system for coal mine underground reservoir coal pillar dam body structure |
| CN114417923A (en) * | 2022-01-13 | 2022-04-29 | 哈尔滨工业大学 | Vehicle induced stress robust correlation mapping method based on convolution noise reduction self-encoder |
| CN116092614A (en) * | 2023-03-06 | 2023-05-09 | 山东大学 | Carbon fiber precursor preparation simulation method based on hybrid neural network |
| CN117116476A (en) * | 2023-07-04 | 2023-11-24 | 中国医学科学院阜外医院 | Downstream task prediction method and device and computer readable storage medium |
| CN117116476B (en) * | 2023-07-04 | 2023-12-19 | 中国医学科学院阜外医院 | Downstream task prediction method and device and computer readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111046080A (en) | Carbon fiber precursor monitoring data preprocessing method based on convolution denoising autoencoder | |
| Cai et al. | Path-level network transformation for efficient architecture search | |
| CN110349146B (en) | Method for constructing fabric defect identification system based on lightweight convolutional neural network | |
| CN112577747A (en) | Rolling bearing fault diagnosis method based on space pooling network | |
| CN110991424A (en) | Fault diagnosis method based on minimum entropy deconvolution and stacking sparse self-encoder | |
| CN111161207B (en) | Integrated convolutional neural network fabric defect classification method | |
| CN109783910B (en) | Structure optimization design method for accelerating by using generation countermeasure network | |
| CN113203566B (en) | Motor bearing fault diagnosis method based on one-dimensional data enhancement and CNN | |
| CN113923104B (en) | Network fault diagnosis method, equipment and storage medium based on wavelet neural network | |
| CN107633272B (en) | DCNN texture defect identification method based on compressed sensing under small sample | |
| CN115204035A (en) | Generator set operation parameter prediction method and device based on multi-scale time sequence data fusion model and storage medium | |
| Yu et al. | CWGAN: Conditional wasserstein generative adversarial nets for fault data generation | |
| CN111666986A (en) | Machine learning-based crayfish grading method | |
| CN110390358A (en) | A kind of deep learning method based on feature clustering | |
| CN115290326A (en) | Rolling bearing fault intelligent diagnosis method | |
| CN112884149A (en) | Deep neural network pruning method and system based on random sensitivity ST-SM | |
| CN115452376A (en) | Bearing fault diagnosis method based on improved lightweight deep convolution neural network | |
| CN110598326A (en) | Well testing interpretation method based on artificial intelligence | |
| CN114386452A (en) | Method for detecting faults of sun wheel of nuclear power circulating water pump | |
| CN108537581B (en) | Energy consumption time series prediction method and device based on GMDH selective combination | |
| CN115660221B (en) | Oil and gas reservoir economic recoverable reserve assessment method and system based on hybrid neural network | |
| CN114235409A (en) | Rolling bearing multi-user cooperative intelligent fault diagnosis method for light weight communication | |
| CN110739030B (en) | Soft measurement method for small sample in ethylene production process | |
| CN112444395A (en) | CMWPE and SaE-ELM based locomotive wheel pair bearing fault diagnosis method | |
| CN112257791A (en) | Classification method of multi-attribute classification tasks based on CNN and PCA |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200421 |