CN108845072B - 4-CBA content dynamic soft measurement method based on convolutional neural network - Google Patents

4-CBA content dynamic soft measurement method based on convolutional neural network Download PDF

Info

Publication number
CN108845072B
CN108845072B CN201810735405.8A CN201810735405A CN108845072B CN 108845072 B CN108845072 B CN 108845072B CN 201810735405 A CN201810735405 A CN 201810735405A CN 108845072 B CN108845072 B CN 108845072B
Authority
CN
China
Prior art keywords
layer
pooling
soft measurement
convolutional
dynamic soft
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
Application number
CN201810735405.8A
Other languages
Chinese (zh)
Other versions
CN108845072A (en
Inventor
刘瑞兰
周鹏
龚梦龙
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Posts and Telecommunications
Original Assignee
Nanjing University of Posts and Telecommunications
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing University of Posts and Telecommunications filed Critical Nanjing University of Posts and Telecommunications
Priority to CN201810735405.8A priority Critical patent/CN108845072B/en
Publication of CN108845072A publication Critical patent/CN108845072A/en
Application granted granted Critical
Publication of CN108845072B publication Critical patent/CN108845072B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N31/00—Investigating or analysing non-biological materials by the use of the chemical methods specified in the subgroup; Apparatus specially adapted for such methods
    • G01N31/005—Investigating or analysing non-biological materials by the use of the chemical methods specified in the subgroup; Apparatus specially adapted for such methods investigating the presence of an element by oxidation
    • 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
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • 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)
  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • General Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Feedback Control In General (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a convolutional neural network-based dynamic soft measurement method for 4-CBA content, which is used for calculating the 4-CBA content generated in a PTA oxidation process, the method firstly constructs a mapping relation between the input and the output of a dynamic soft measurement model based on a convolutional neural network, takes a time sequence data block of a relevant measurable variable in the PTA oxidation process as the input of the dynamic soft measurement model, and takes the 4-CBA as the output; inputting a time sequence data block into a convolutional neural network in which a convolutional layer and a pooling layer are alternately distributed, wherein the convolutional layer and the pooling layer are both 2 layers, the first layer of pooling adopts one-dimensional maximum pooling to extract features after convolution, the second layer of pooling adopts maximum pooling with the same size as a feature map output by the convolutional layer for sampling, the output of the last layer of pooling is calculated by using a linear function to obtain an output result, and the result is compared with 4-CBA analysis data and parameters are updated; the dynamic soft measurement model is simple and easy to realize, and the measurement precision of the model is improved.

Description

4-CBA content dynamic soft measurement method based on convolutional neural network
Technical Field
The invention belongs to the field of chemical engineering, and particularly relates to a 4-CBA content dynamic soft measurement method based on a convolutional neural network.
Background
The convolutional neural network is a classical feedforward neural network, is mainly proposed by the concept of receptive field in biology, and is an important model in deep learning, each layer of the convolutional neural network consists of a convolutional layer and a subsequent downsampling layer, and the convolutional layer performs convolution operation on the output of an upper layer by using a convolutional core. Compared with a full-connection method, the structure is more consistent with the working mode of biological neurons, and network parameters are reduced, so that overfitting is inhibited, and the training speed is accelerated; and the downsampling carries out statistical calculation on the convolution result, so that the obtained features have certain translation invariance and rotation invariance.
Purified Terephthalic Acid (PTA) is an important raw material in the production of polyesters and chemical industry. The poly-p-carboxybenzaldehyde (4-CBA) is one of the main byproducts in the oxidation reaction of p-xylene (PX), and is also a main impurity and an important control index in a PTA product. However, in actual production, the content of 4-CBA is often obtained by sampling and offline analysis, but the delay time of offline analysis is often several hours long, and the sampling times are few, so that the control requirement cannot be met. In response to this problem, soft measurement techniques are required to estimate the 4-CBA content on-line.
At present, most of the existing 4-CBA content soft measurement models are static soft measurement models, and a supervised shallow algorithm is adopted. Algorithms such as PCA, PLS and BP are mature 4-CBA soft measurement models, but the algorithms are shallow-layer structures, the method is limited in the representation capability of complex functions under the condition of limited samples and computing units, the generalization capability of the algorithms is limited, the algorithms are easy to fall into local optimum, and the methods are mostly based on static models, so that when the working condition changes, the model precision is reduced, and the robustness is poor.
Disclosure of Invention
The invention mainly aims to provide a convolutional neural network-based dynamic soft measurement method for 4-CBA content, which uses a time series data block related to measurable variables in a PTA oxidation process as model input, introduces dynamic characteristics of the variables in a production process into a model to obtain an efficient dynamic soft measurement model for the 4-CBA content, effectively solves the problems of low measurement precision and poor robustness of the existing static soft measurement model, and has the following specific technical scheme:
a dynamic soft measurement method of 4-CBA content based on a convolutional neural network is used for calculating the 4-CBA content generated in a PTA oxidation process, firstly, a dynamic soft measurement model is constructed based on the convolutional neural network, the dynamic soft measurement model comprises a first convolutional layer and a second convolutional layer, a first pooling layer and a second pooling layer are respectively a pooling layer and an output layer, and the convolutional layers and the pooling layers are alternately distributed; the dynamic soft measurement method for the content of the 4-CBA comprises the following steps:
s100: selecting related measurable variables in a specified amount of PTA oxidation processes as input variables of the dynamic soft measurement model, and using the content of 4-CBA as the output of the dynamic soft measurement model;
s200: acquiring m groups of time sequence data blocks corresponding to the input and 4-CBA content data corresponding to the output as initial training samples, and standardizing the initial training samples;
s300: setting an iteration number t, inputting the sample data of the initialized training sample after standardized processing into a dynamic soft measurement model for convolution operation, and performing sampling operation on the feature graph after convolution through pooling;
s400: dropout is used in the last pooling operation, pooling output of the second pooling layer is calculated through a linear function, a calculation result is compared with analysis data of 4-CBA content, and a dynamic soft measurement model is updated through back propagation;
s500: and recording the current iteration number as k, if k is less than t, repeating the steps S3 and S4, otherwise, storing the parameters of the dynamic soft measurement model to obtain the dynamic soft measurement model, and realizing the soft measurement operation of the 4-CBA content by the finally obtained dynamic soft measurement model.
Further, the input variables include CO2 content generated by the reaction during PTA oxidation, mix tank feed flow, reactor level, reactor temperature, and first crystallizer temperature constraints.
Further, if the time-series data block is x, x can be represented by the formula
Figure GDA0002758139360000031
Where T is the transition time length, τ1＝150min，τ2＝144min，τ3＝102min，τ4＝114min，τ5＝18min；x1For CO formed by the reaction2Content, x2For the feed rate of the mixing tank, x3Is the reactor level, x4Is the reactor temperature, x5Is the first crystallizer temperature constraint.
Furthermore, the two convolutional layers both adopt 3 × 3 convolutional kernels, and the number of the convolutional kernels of the convolutional layers is designed by adopting a dimension reduction method, wherein the number of the convolutional kernels of the first convolutional layer is 64, and the number of the convolutional kernels of the second convolutional layer is 32; the first pooling layer adopts one-dimensional maximum pooling, the pooling size is 6 multiplied by 1, the second pooling layer adopts maximum pooling equal to the size of a feature map output by the second convolution layer, and the output layer calculates an output result by using a linear function.
Further, the probability of dropout being selected to be 0.3 disconnects between neurons.
According to the method, a dynamic soft measurement model is established through a convolutional neural network, time series data of historical operation corresponding to input variables are used as input of the model, dynamic characteristics of a sample are better extracted through sampling weight sharing and local receptive fields, redundant information of the sample is removed through a method of reducing the number dimension of convolutional kernels, in addition, the last layer of pooling uses the largest pooling which is equal to the size of a characteristic graph output by a convolutional layer to replace a full-connection layer, the number of parameters is reduced, and overfitting can be well reduced; the method provided by the invention introduces dynamic characteristics in industrial production into the model, and improves the prediction precision of the model.
Drawings
FIG. 1 is a flow chart of the method for dynamically measuring the content of 4-CBA based on the convolutional neural network according to the present invention;
FIG. 2 is a block diagram illustration of a convolutional neural network in accordance with the present invention.
In the figure: x is model input; c1 first convolutional layer; s1 max pooling layer, C2 second convolutional layer; s2 global maximum pooling layers, W1, …, W32 are weights of the linear output layers; and Y is the model output.
Detailed Description
In order to make the technical solutions of the present invention better understood, 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.
Referring to fig. 1 and fig. 2, in the embodiment of the present invention, a dynamic soft measurement method of 4-CBA content based on a convolutional neural network is provided, which is used for calculating the 4-CBA content generated in a PTA oxidation process, first, a dynamic soft measurement model is constructed based on the convolutional neural network, the dynamic soft measurement model includes two convolutional layers, namely a first convolutional layer and a second convolutional layer, two pooling layers, namely a first pooling layer and a second pooling layer, and an output layer, and the convolutional layers and the pooling layers are alternately distributed; the dynamic soft measurement method for the content of the 4-CBA comprises the following steps: s100: selecting related measurable variables in a specified amount of PTA oxidation processes as input variables of the dynamic soft measurement model, and using the content of 4-CBA as the output of the dynamic soft measurement model; s200: acquiring m groups of time sequence data blocks corresponding to the input and 4-CBA content data corresponding to the output as initial training samples, and standardizing the initial training samples; s300: setting an iteration number t, inputting the sample data of the initialized training sample after standardized processing into a dynamic soft measurement model for convolution operation, and performing sampling operation on the feature graph after convolution through pooling; s400: dropout is used in the last pooling operation, pooling output of the second pooling layer is calculated through a linear function, a calculation result is compared with analysis data of 4-CBA content, and a dynamic soft measurement model is updated through back propagation; s500: and recording the current iteration number as k, if k is less than t, repeating the steps S3 and S4, otherwise, storing the parameters of the dynamic soft measurement model to obtain the dynamic soft measurement model, and realizing the soft measurement operation of the 4-CBA content by the finally obtained dynamic soft measurement model.
In this embodiment, in step S1, the input variables include five variables of CO2 content generated by the reaction in the PTA oxidation process, feed flow rate of the mixing tank, reactor liquid level, reactor temperature and first crystallizer temperature constraint as inputs of the convolutional neural network-based 4-CBA content dynamic soft measurement method of the present invention; of course, the present invention is not limited or fixed, and the specific input variables may be selected according to actual situations.
In one embodiment, assuming the dynamic soft-measurement input, i.e., the time series data block is x, x can be represented by the formula
Figure GDA0002758139360000051
Where T is the length of the transition time, length is 30, τ1＝150min，τ2＝144min，τ3＝102min，τ4＝114min，τ5＝18min；x1For CO formed by the reaction2Content, x3For the feed rate of the mixing tank, x3Is the reactor level, x4Is the reactor temperature, x5Is the first crystallizer temperature constraint.
In the embodiment of the invention, two convolutional layers both adopt 3 × 3 convolutional kernels, and the number of the convolutional kernels of the convolutional layers is designed by adopting a dimension reduction method, wherein the number of the convolutional kernels of the first convolutional layer is 64, and the number of the convolutional kernels of the second convolutional layer is 32; the first pooling layer adopts one-dimensional maximum pooling, the pooling size is 6 multiplied by 1, the second pooling layer adopts maximum pooling equal to the size of a feature map output by the second convolution layer, and the output layer calculates an output result by using a linear function; specifically, with reference to fig. 2, the C1 layer is a convolution layer having 64 convolutional layers with a size of 3 × 3, the input sample is convolved by the C1 layer to obtain 64 feature maps with a size of 30 × 5, the feature map obtained by the C1 layer is used as the input of the S1 layer, the S1 layer is maximally pooled at 6 × 1 to obtain 64 feature maps with a size of 5 × 5 and is used as the input of the C2 layer, the C2 layer is a convolutional layer with a size of 3 × 3, the output of the S1 layer is maximally pooled by the C1 layer to obtain 32 feature maps with a size of 5 × 5, the feature map obtained from the C2 layer is connected to the S2 layer, the size of the S2 is equal to the size of the feature map obtained from the C2 layer, the obtained features are maximally pooled to obtain 32 features, the obtained features are calculated by the linear output layer to obtain training results, and are compared with the analysis data of 4-CBA content and the model parameters are updated by using back propagation; wherein, the learning step length of the convolutional neural network is 0.01, and the training error of the convolutional neural network output layer is adopted
Figure GDA0002758139360000052
Wherein y is a desired value and y' is an output value; also to prevent overfitting, dropout was used to disconnect neurons with a probability of 0.3.
According to the method, a dynamic soft measurement model is established through a convolutional neural network, time series data of historical operation corresponding to input variables are used as input of the model, dynamic characteristics of a sample are better extracted through sampling weight sharing and local receptive fields, redundant information of the sample is removed through a method of reducing the number dimension of convolutional kernels, in addition, the last layer of pooling uses the largest pooling which is equal to the size of a characteristic graph output by a convolutional layer to replace a full-connection layer, the number of parameters is reduced, and overfitting can be well reduced; the method provided by the invention introduces dynamic characteristics in industrial production into the model, and improves the prediction precision of the model.
Although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that modifications may be made to the embodiments described in the foregoing detailed description, or equivalent changes may be made in some of the features of the embodiments described above. All equivalent structures made by using the contents of the specification and the attached drawings of the invention can be directly or indirectly applied to other related technical fields, and are also within the protection scope of the patent of the invention.

Claims (5)

1. A dynamic soft measurement method of 4-CBA content based on a convolutional neural network is used for calculating the 4-CBA content generated in a PTA oxidation process, and is characterized in that a dynamic soft measurement model is firstly constructed based on the convolutional neural network, the dynamic soft measurement model comprises two convolutional layers of a first convolutional layer and a second convolutional layer, two pooling layers of the first pooling layer and the second pooling layer and an output layer, and the convolutional layers and the pooling layers are alternately distributed; the dynamic soft measurement method for the content of the 4-CBA comprises the following steps:
s100: selecting related measurable variables in a specified amount of PTA oxidation processes as input variables of the dynamic soft measurement model, and using the content of 4-CBA as the output of the dynamic soft measurement model;
s200: acquiring time sequence data blocks corresponding to m groups of input variables and 4-CBA content data output correspondingly as initial training samples, and standardizing the initial training samples;
s300: setting an iteration number t, inputting the sample data of the initialized training sample after standardized processing into a dynamic soft measurement model for convolution operation, and performing sampling operation on the feature graph after convolution through pooling;
s400: dropout is used in the last pooling operation, pooling output of the second pooling layer is calculated through a linear function, a calculation result is compared with analysis data of 4-CBA content, and a dynamic soft measurement model is updated through back propagation;
s500: and recording the current iteration number as k, if k is less than t, repeating the steps S300 and S400, otherwise, storing the parameters of the dynamic soft measurement model to obtain the dynamic soft measurement model, and realizing the soft measurement operation of the 4-CBA content by the finally obtained dynamic soft measurement model.
2. The convolutional neural network based dynamic soft measurement method of 4-CBA content as claimed in claim 1, wherein the input variables include CO2 content generated by reaction during PTA oxidation, mixing tank feed flow, reactor liquid level, reactor temperature and first crystallizer temperature constraint.
3. The convolutional neural network-based dynamic soft measurement method for 4-CBA content as claimed in claim 1, wherein if the time series data block is x, x can be represented by the formula
Figure FDA0002758139350000021
Where T is the transition time length, τ1＝150min，τ2＝144min，τ3＝102min，τ4＝114min，τ5＝18min；x1For CO formed by the reaction2Content, x2For the feed rate of the mixing tank, x3Is the reactor level, x4Is the reactor temperature, x5Is the first crystallizer temperature constraint.
4. The convolutional neural network-based dynamic soft measurement method of 4-CBA content, as claimed in claim 1, wherein both convolutional layers use 3 × 3 convolutional kernels, and the number of convolutional kernels of the convolutional layers is designed by using a dimension reduction method, wherein the number of convolutional kernels of the first convolutional layer is 64, and the number of convolutional kernels of the second convolutional layer is 32; the first pooling layer adopts one-dimensional maximum pooling, the pooling size is 6 multiplied by 1, the second pooling layer adopts maximum pooling equal to the size of a feature map output by the second convolution layer, and the output layer calculates an output result by using a linear function.
5. The convolutional neural network-based dynamic soft measurement method of 4-CBA content as claimed in claim 1, wherein dropout is selected to have a probability of 0.3 to disconnect neurons.
CN201810735405.8A 2018-07-06 2018-07-06 4-CBA content dynamic soft measurement method based on convolutional neural network Active CN108845072B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810735405.8A CN108845072B (en) 2018-07-06 2018-07-06 4-CBA content dynamic soft measurement method based on convolutional neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810735405.8A CN108845072B (en) 2018-07-06 2018-07-06 4-CBA content dynamic soft measurement method based on convolutional neural network

Publications (2)

Publication Number Publication Date
CN108845072A CN108845072A (en) 2018-11-20
CN108845072B true CN108845072B (en) 2020-12-15

Family

ID=64201683

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810735405.8A Active CN108845072B (en) 2018-07-06 2018-07-06 4-CBA content dynamic soft measurement method based on convolutional neural network

Country Status (1)

Country Link
CN (1) CN108845072B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110189800B (en) * 2019-05-06 2021-03-30 浙江大学 Furnace oxygen content soft measurement modeling method based on multi-granularity cascade cyclic neural network
CN110378044B (en) * 2019-07-23 2021-06-11 燕山大学 Multi-time scale convolution neural network soft measurement method based on attention mechanism
CN113190797A (en) * 2021-04-18 2021-07-30 宁波大学科学技术学院 PTA device gross error discrimination method based on online rolling discrimination feature analysis

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105825509A (en) * 2016-03-17 2016-08-03 电子科技大学 Cerebral vessel segmentation method based on 3D convolutional neural network
US9830526B1 (en) * 2016-05-26 2017-11-28 Adobe Systems Incorporated Generating image features based on robust feature-learning
US11321613B2 (en) * 2016-11-17 2022-05-03 Irida Labs S.A. Parsimonious inference on convolutional neural networks
CN106599941A (en) * 2016-12-12 2017-04-26 西安电子科技大学 Method for identifying handwritten numbers based on convolutional neural network and support vector machine

Also Published As

Publication number Publication date
CN108845072A (en) 2018-11-20

Similar Documents

Publication Publication Date Title
Yuan et al. Hierarchical quality-relevant feature representation for soft sensor modeling: A novel deep learning strategy
CN110689965B (en) Drug target affinity prediction method based on deep learning
CN108845072B (en) 4-CBA content dynamic soft measurement method based on convolutional neural network
CN107451101B (en) Method for predicting concentration of butane at bottom of debutanizer by hierarchical integrated Gaussian process regression soft measurement modeling
CN111640510A (en) Disease prognosis prediction system based on deep semi-supervised multitask learning survival analysis
CN111079836B (en) Process data fault classification method based on pseudo label method and weak supervised learning
CN108920888B (en) Continuous stirred tank type reaction process identification method based on deep neural network
Taylan et al. A new approach to multivariate adaptive regression splines by using Tikhonov regularization and continuous optimization
CN110516818A (en) A kind of high dimensional data prediction technique based on integrated study technology
Wang et al. Diabetes Risk Analysis Based on Machine Learning LASSO Regression Model
CN113033090B (en) Push model training method, data push device and storage medium
CN115618196A (en) Transformer anomaly detection method based on space-time characteristics
Zhang et al. Online quality prediction for cobalt oxalate synthesis process using least squares support vector regression approach with dual updating
CN112733692B (en) Fault prediction method and device based on integrated hybrid model and related equipment
CN104504232B (en) A kind of flexible measurement method of the 4 CBA contents based on Sparse least squares support vector machine
CN110221540A (en) Continuous-stirring reactor system control method based on Hammerstein model
CN112966770A (en) Fault prediction method and device based on integrated hybrid model and related equipment
CN115034504B (en) Cutter wear state prediction system and method based on cloud edge cooperative training
CN114297795B (en) PR-Trans-based mechanical equipment residual life prediction method
Zhang et al. Active learning strategy for online prediction of particle size distribution in cobalt oxalate synthesis process
Vaitkus et al. Soft-sensors based on black-box models for bioreactors monitoring and state estimation
Chaikivskyi et al. Development of Neural Network for Cyclohexane Oxidation Data Processing
CN112364527A (en) Debutanizer soft measurement modeling method based on ALIESN online learning algorithm
Chaikivskyi et al. Development of Artificial Neural Networks for Cyclohexane Oxidation Data Analysis
Wei et al. Real-time forecasting of subsurface inclusion defects for continuous casting slabs: A data-driven comparative study

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