CN117289594A - Method and system for controlling water content of outlet of cut tobacco drying based on mode discrimination - Google Patents
Method and system for controlling water content of outlet of cut tobacco drying based on mode discrimination Download PDFInfo
- Publication number
- CN117289594A CN117289594A CN202311193095.9A CN202311193095A CN117289594A CN 117289594 A CN117289594 A CN 117289594A CN 202311193095 A CN202311193095 A CN 202311193095A CN 117289594 A CN117289594 A CN 117289594A
- Authority
- CN
- China
- Prior art keywords
- data
- monitoring data
- time
- outlet
- water content
- 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
- 238000000034 method Methods 0.000 title claims abstract description 62
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 title claims abstract description 52
- 238000001035 drying Methods 0.000 title claims abstract description 33
- 241000208125 Nicotiana Species 0.000 title claims abstract description 26
- 235000002637 Nicotiana tabacum Nutrition 0.000 title claims abstract description 26
- 238000012549 training Methods 0.000 claims abstract description 38
- 238000001212 derivatisation Methods 0.000 claims abstract description 24
- 238000007781 pre-processing Methods 0.000 claims abstract description 19
- 238000003860 storage Methods 0.000 claims abstract description 10
- 238000012544 monitoring process Methods 0.000 claims description 101
- 239000011159 matrix material Substances 0.000 claims description 39
- 230000002159 abnormal effect Effects 0.000 claims description 28
- 230000008569 process Effects 0.000 claims description 18
- 230000009467 reduction Effects 0.000 claims description 17
- 230000006870 function Effects 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 10
- 238000009826 distribution Methods 0.000 claims description 8
- 238000004519 manufacturing process Methods 0.000 claims description 8
- 239000000463 material Substances 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000013528 artificial neural network Methods 0.000 claims description 6
- 238000001514 detection method Methods 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 6
- 230000005856 abnormality Effects 0.000 claims description 5
- 239000000567 combustion gas Substances 0.000 claims description 5
- 238000007637 random forest analysis Methods 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 4
- 230000001965 increasing effect Effects 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 230000006835 compression Effects 0.000 claims description 3
- 238000007906 compression Methods 0.000 claims description 3
- 238000013500 data storage Methods 0.000 claims description 3
- 239000007789 gas Substances 0.000 claims description 3
- 230000015556 catabolic process Effects 0.000 claims description 2
- 238000006731 degradation reaction Methods 0.000 claims description 2
- 238000005065 mining Methods 0.000 claims description 2
- 230000007613 environmental effect Effects 0.000 abstract description 6
- 238000013461 design Methods 0.000 abstract description 4
- 230000002708 enhancing effect Effects 0.000 abstract description 2
- 230000006872 improvement Effects 0.000 description 8
- 238000010586 diagram Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 238000004804 winding Methods 0.000 description 2
- 238000010923 batch production Methods 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000018044 dehydration Effects 0.000 description 1
- 238000006297 dehydration reaction Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000013072 incoming material Substances 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- A—HUMAN NECESSITIES
- A24—TOBACCO; CIGARS; CIGARETTES; SIMULATED SMOKING DEVICES; SMOKERS' REQUISITES
- A24B—MANUFACTURE OR PREPARATION OF TOBACCO FOR SMOKING OR CHEWING; TOBACCO; SNUFF
- A24B3/00—Preparing tobacco in the factory
- A24B3/04—Humidifying or drying tobacco bunches or cut tobacco
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B11/00—Automatic controllers
- G05B11/01—Automatic controllers electric
- G05B11/36—Automatic controllers electric with provision for obtaining particular characteristics, e.g. proportional, integral, differential
- G05B11/42—Automatic controllers electric with provision for obtaining particular characteristics, e.g. proportional, integral, differential for obtaining a characteristic which is both proportional and time-dependent, e.g. P. I., P. I. D.
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
- G06F18/15—Statistical pre-processing, e.g. techniques for normalisation or restoring missing data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/27—Regression, e.g. linear or logistic regression
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- 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
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06N3/0985—Hyperparameter optimisation; Meta-learning; Learning-to-learn
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2123/00—Data types
- G06F2123/02—Data types in the time domain, e.g. time-series data
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Probability & Statistics with Applications (AREA)
- Medical Informatics (AREA)
- Automation & Control Theory (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method for controlling the water content of a cut tobacco outlet based on mode discrimination, which comprises the following specific steps: A. data acquisition and storage; B. preprocessing data; C. feature cross-derivatization and enhancement; D. merging attention time convolution networks; E. the attention time convolution network training is integrated, so that the intelligent adjustment of control parameters is realized, and the control of the water content of an outlet is realized. On the basis of data preprocessing and characteristic engineering, the invention designs a fused attention time convolution network to realize mode discrimination of the tobacco drying process, and intelligently adjusts control parameters, thereby realizing control of the outlet moisture content, being capable of accurately grasping equipment state, the tobacco drying process and environmental influence, timely adjusting the control parameters, having no time lag and higher adjustment rationality compared with PID control and manual adjustment, and further enhancing the stability and accuracy of the tobacco outlet moisture content.
Description
Technical Field
The invention relates to the technical field of tobacco shred drying, in particular to a method and a system for controlling the water content of a shred drying outlet based on mode discrimination.
Background
The tobacco shred baking is an important link in the tobacco processing process, and the outlet water content is a key index for evaluating the tobacco shred quality. The tobacco shred drying link mainly enables the moisture content of tobacco shreds to meet the winding requirement through drying and dehydration, improves the tobacco shred expansion effect and reduces the winding consumption. The process of drying the cut tobacco is greatly influenced by factors such as temperature, humidity, the quality of the preamble process and the like, and the control parameters are complex, so that the water content of an outlet of the cut tobacco is difficult to control stably and accurately.
At present, most cut-tobacco drier adopts PID control and relies on operators to manually adjust parameters such as incoming material flow according to experience, but the adjustment lag of the methods is very strong, the intervention can be performed only under the condition that the outlet water content deviates from a standard value, and a long period of time is required from entering the cut-tobacco drier to finishing the cut-tobacco drying, so that the lag intervention can cause unstable quality of the cut-tobacco drying and the problem that the outlet water content continuously fluctuates near the standard value.
In recent years, deep learning techniques typified by neural networks have rapidly developed, and have also been widely used in industrial environments, such as application of image processing techniques to defect detection, application of time-series prediction techniques to equipment remaining life prediction, and the like. The deep learning model has strong learning ability on data, and can simulate various industrial scenes through the data.
Therefore, a method for combining the control of the moisture content of the tobacco cut tobacco outlet with a neural network is needed to improve the accuracy and stability of the moisture content of the tobacco cut tobacco outlet.
Disclosure of Invention
In order to solve the technical problems of low accuracy and poor stability of the moisture content of tobacco cut-tobacco outlet in the prior art, the invention aims to provide a cut-tobacco outlet moisture content control method and system based on mode discrimination, which are used for monitoring data acquisition in the cut-tobacco drying process, designing a fused attention time convolution network on the basis of data preprocessing and characteristic engineering, carrying out omnibearing monitoring on the cut-tobacco drying process in a data driving mode, autonomously judging the adjustment time and adjustment quantity of each control parameter, forming the mode discrimination cut-tobacco outlet moisture content control method, and realizing the accuracy and stability of the outlet moisture content by accurately grasping the equipment state, the cut-tobacco drying process, the environmental influence and the like through fused attention, intelligently adjusting the control parameters.
The invention solves the problems by the following technical proposal:
a method for controlling the water content of a cut tobacco outlet based on mode discrimination comprises the following specific steps:
A. data acquisition and storage
The method comprises the steps of monitoring data acquisition in real time, wherein the acquired data comprises monitoring data and control parameters;
recording the specific values of the monitoring data and the control parameters of each batch according to the production plan, forming historical monitoring data of the finished batch, and forming real-time monitoring data of the current production batch;
storing data, namely storing monitoring data by using a MySQL database;
B. data preprocessing
Abnormal moment identification, namely eliminating abnormal values causing data abnormality by locating the abnormal moment in the data acquisition process;
filling the missing value, filling the missing value and the deleted monitoring data in abnormal moment identification in the data acquisition process, training a random forest regression model by taking the complete historical monitoring data as a training set, and predicting and filling the missing value and the deleted detection data;
normalizing the data, and normalizing the monitoring data;
C. feature cross-derivatization and enhancement
Feature cross derivatization, namely performing feature cross derivatization on the monitoring data after data preprocessing, mining internal connection among the monitoring data, specifically encoding a nonlinear rule in a feature space through multi-feature Cartesian product, and generating a synthesized feature, thereby increasing the dimension of the feature;
feature enhancement, namely performing degradation and enhancement on derived features, specifically designing a noise reduction self-encoder for the characteristics of the monitored data, and performing depth fusion and enhancement on the cross derived features through the noise reduction self-encoder;
D. fused attention time convolution network
Designing a structure of a convolution network integrating attention time, realizing mode discrimination based on monitoring data, and intelligently identifying adjustment time and adjustment quantity of each parameter;
E. fused attention time convolutional network training
Constructing training samples and labels, dividing historical monitoring data into training samples according to a certain time after preprocessing, feature cross derivatization and enhancement, and calculating the adjustment quantity of each control parameter at the end of a certain time of each training sample to serve as the labels, wherein one training sample corresponds to a plurality of labels;
then training the fused attention time convolution network according to different labels respectively to obtain mode discrimination models corresponding to a plurality of labels;
the real-time monitoring data of the current batch is input into a mode discrimination model, so that the intelligent adjustment of control parameters is realized, and the control of the water content of the outlet is realized.
As a further improvement of the present invention, the monitoring data includes at least outlet water content, outlet temperature, inlet water content, inlet temperature, return gas temperature, air temperature, moisture removal negative pressure, ambient temperature and ambient humidity; the control parameters at least comprise material flow, drying temperature, combustion gas flow and steam flow; the total of the monitoring data and the control parameters is not less than 13, that is to say the dimension of the monitoring data is at least 13.
As a further improvement of the invention, the MySQL database divides the collected data by date and stores the collected data on different dates into different tables.
As a further improvement of the invention, the MySQL database is used for storing the monitoring data, and the specific method comprises the following steps: and acquiring and writing the monitoring data every second, forming monitoring data with the time interval of 1s in a MySQL database, assigning unique ids for each batch for distinguishing, and simultaneously creating a group_record table to record all batch information, wherein the group_record table at least comprises batch id, start time, end time and material license plate information.
As a further improvement of the invention, the abnormal time is identified, and the abnormal value causing the data abnormality is removed by locating the abnormal time in the data acquisition process;
the abnormal moment is positioned by means of multi-element Gaussian distribution detection, and the method specifically comprises the following steps:
assume that the collected monitoring data set is:
wherein n is the dimension of the monitoring data;
the mean vector of n dimensions is calculated as:
wherein E (x) i ) The average value of the i-th dimension monitoring data is m, and the m is the number of certain monitoring data;
thereby calculating the monitoring dataIs a covariance matrix of (a):
then the data is monitoredThe distribution probability of (2) is:
probability of distribution of the monitored data for each instant, ifThe time is marked as an abnormal time, and the monitoring data at the time is deleted.
As a further improvement of the invention, the data normalization, the normalization processing is performed on the monitored data, and the normalization processing is realized by using the maximum and minimum values, and the formula is utilized:
wherein x is * For normalized monitoring data, x is the original monitoring data, min (x) is the minimum value of the original monitoring data, and max (x) is the maximum value of the original monitoring data.
As a further improvement of the invention, the noise reduction self-encoder consists of three parts, namely a random noise introduction part, an encoding module and a decoding module; the noise reduction self-Encoder is a self-supervision learning model which trains the original data as a label, encodes the original input added with noise into hidden layer output in an Encoder, and reconstructs the hidden output into the original input in the Decoder so as to realize the dimensional compression and enhancement of the characteristics; the hidden layer output in the noise reduction self-encoder is the final characteristic after enhancement.
As a further improvement of the present invention, the fused attention time convolution network includes a self-attention portion and a time convolution portion, provided that the real-time monitoring data is transformed into the data after cross-derivatization and feature enhancementWherein N is the number of features, and T is the time span;
in the self-attention part, three transformation matrices w are defined q ,w k ,w v Performing linear transformation on X for three times to obtain a derivative matrix of X, and respectively marking the derivative matrix as a query matrix Q, a key matrix K and a value matrix V; then, updating the value matrix by using the query matrix and the key matrix, so that the model selectively focuses on information of different dimensions and different moments, and simultaneously, corresponding weights are distributed for each position of the input X, and the self-focusing calculation result is as follows:
R=σ[QK T ]V;
wherein σ is a nonlinear activation function that maps the value range between (0, 1);
sigma adopts a Sigmoid function:
the Sigmoid function is most sensitive at the position of 0.5, and can well distinguish the part needing to be reserved or inhibited in the value matrix;
in the time convolution part, convolution operation is commonly used for processing an image, the size of a convolution kernel in a characteristic dimension is set to be 1, and the size of a time dimension is set to be t, so that the convolution operation is only performed in the time dimension;
specifically, the convolution kernel is:
U=[u 1 ,u 2 ,…,u t ];
the self-attention calculation result is convolved to obtain:
Y=R⊙U;
wherein, as follows;
finally, the output Y of the time convolution is transmitted into a fully-connected neural network, and N characteristics are fused and transformed into one-dimensional data:
F=w f Y+b;
wherein w is f And b is an offset, and F is the final output of the fusion attention time convolution network.
As a further improvement of the invention, the network trains, and then trains the fused attention time convolution network according to different labels respectively to obtain a mode discrimination model corresponding to a plurality of labels;
updating model parameters by adopting an Adam optimizer to finish training the model, wherein the updating process is as follows:
m t =δ 1 ·m t-1 +(1-δ 1 )·g t ;
wherein g t For the gradient of the objective function at time t, m t For first order matrix estimation, n t Delta for second order matrix estimation 1 And delta 2 Exponential decay Rate, gamma, estimated for matrix t For the parameters to be updated, α is the learning rate and ε is a small constant added to maintain data stability.
Meanwhile, the technical scheme of the invention further comprises the following steps:
the control system for the water content of the outlet of the drying wire based on the mode discrimination is used for realizing the control method for the water content of the outlet of the drying wire based on the mode discrimination, and comprises a data acquisition and storage module, a data preprocessing module, a characteristic cross derivatization and enhancement module, a fused attention time convolution network module and a fused attention time convolution network training module, wherein the data acquisition and storage module is used for monitoring data acquisition and data storage in real time, the data preprocessing module is used for abnormal moment identification, missing value filling and data normalization processing, the characteristic cross derivatization and enhancement module is used for characteristic cross derivatization and characteristic enhancement, the fused attention time convolution network module is used for designing a fused attention time convolution network structure, and the fused attention time convolution network training module is used for constructing training samples and labels and training a fused attention time convolution network.
Compared with the prior art, the invention has the following advantages:
on the basis of data preprocessing and characteristic engineering, the invention designs a fused attention time convolution network to realize mode discrimination of the silk drying process, and intelligently adjusts control parameters so as to realize control of outlet water content.
The control method disclosed by the invention can accurately grasp the equipment state, the drying process and the environmental influence, and timely adjust the control parameters, compared with PID control and manual adjustment, the control method has the advantages of no time lag, higher adjustment rationality and stronger stability and accuracy of the outlet water content.
Drawings
FIG. 1 is a flow chart of a method for controlling the water content of an outlet of a cut tobacco dryer based on mode discrimination in an embodiment of the invention;
FIG. 2 is a schematic diagram of an intermediate noise reduction self-encoder according to an embodiment of the present invention;
FIG. 3 is a schematic diagram showing the effect of the outlet water content under PID/manual adjustment in the embodiment of the invention;
fig. 4 is a schematic diagram of an effect of outlet water content under adjustment of a control method of outlet water content of a baked wire based on mode discrimination in the embodiment of the invention;
fig. 5 is a block diagram of a control system for controlling the water content of an outlet of a cut tobacco dryer based on mode discrimination in an embodiment of the invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
In the case of example 1,
referring to fig. 1, a method for controlling the water content of a cut tobacco drying outlet based on mode discrimination specifically comprises the following steps:
A. data acquisition and storage
A1. Monitoring data acquisition in real time;
the collected data comprise monitoring data and control parameters, in this embodiment, specifically, the monitoring data comprise outlet water content, outlet temperature, inlet water content, inlet temperature, return gas temperature, air temperature, moisture removal negative pressure, ambient temperature and ambient humidity; the control parameters include material flow, drying temperature, combustion gas flow and steam flow. The total of the monitoring data and the control parameters is 13, that is to say the dimension of the monitoring data is 13.
Recording specific values of each batch of monitoring data and control parameters according to the production plan, forming historical monitoring data of finished batches, and forming real-time monitoring data of the current production batch.
A2. Storing data;
the MySQL database is used for storing the monitoring data, is the most popular relational database, and stores the data with different dates into different tables by dividing the data according to the dates, so that the flexibility and the running speed of the data are improved.
Specifically, monitoring data are collected and written in each second, monitoring data with a time interval of 1s are formed in a MySQL database, unique ids are assigned to each batch for distinguishing, and a group_record table is created to record all batch information, including information such as batch id, starting time, ending time, material license plate and the like.
B. Data preprocessing
B1. Identifying abnormal time;
in the process of monitoring data acquisition, factors such as equipment instability, environmental disturbance, improper operation and the like can cause data abnormality, and the influence of abnormal values on a model is large, so that the abnormal values need to be removed. In this embodiment, the abnormal value found is deleted by locating the abnormal time by means of multivariate gaussian distribution detection.
Specifically, assume that the collected monitoring data set is:
where n is the dimension of the monitoring data, in this embodiment, the monitoring data is 13 dimensions.
The mean vector of n dimensions is calculated as:
wherein E (x) i ) And m is the number of certain monitoring data, wherein m is the mean value of the i-th dimension monitoring data.
Thereby calculating the monitoring dataIs a covariance matrix of (a):
then the data is monitoredThe distribution probability of (2) is:
for each moment in time, if it isThe time is marked as an abnormal time, and the monitoring data at the time is deleted.
B2. Filling up the missing value;
aiming at the missing value in the data acquisition process and the deleted monitoring data in the abnormal moment, in the embodiment, a random forest regression method is adopted to fill the missing value and the deleted monitoring data, the complete historical monitoring data is used as a training set to train a random forest regression model under cross verification, and the missing value is predicted and filled.
Specifically, the random forest is a Bagging integration algorithm, and in this embodiment, an average value of prediction results of the base estimator is used as a result of the integration estimator. For historical monitoring data, a 25 class tree model is built as a base evaluator.
B3. Normalizing the data;
in order to eliminate the influence of different dimensions, the model convergence is quickened, and the monitoring data is normalized.
Specifically, in this embodiment, the maximum and minimum normalization is used, and the formula is as follows:
wherein x is * For normalized monitoring data, x is the original monitoring data.
C. Feature cross-derivatization and enhancement
C1. Feature cross-derivatization;
in actual production, the monitoring data often has the characteristics of strong coupling, low value density and the like, and better effect cannot be obtained by directly modeling the original monitoring data. Thus, feature cross-derivatization is performed on the monitored data, and the inherent links between the data are mined. During the process of drying the shreds, a single feature often cannot be directly associated with a change in the outlet moisture content, but a combination of multiple features may become a key factor in the outlet moisture content change.
Specifically, the cross dimension rise is carried out on the monitoring data, the fitting capacity after modeling is improved, and the base vector has higher expressive capacity. In this embodiment, the non-linear law in the feature space is encoded by taking the cartesian product of multiple features, resulting in a composite feature, thereby increasing the dimension of the feature.
C2. Enhancing the characteristics;
after the feature dimension is increased, the problems of strong sparsity, high calculation cost and the like are inevitably caused, and meanwhile, not every derivative feature has a strong correlation with the water content of the outlet, so that the derivative feature is required to be subjected to dimension reduction and enhancement.
Specifically, in this embodiment, a noise reduction self-encoder is designed for the characteristics of the monitored data, and depth fusion and enhancement are performed on the cross-derived features. The inter noise reduction self-Encoder consists of three parts, namely a random noise introduction part, an encoding module (Encoder) part and a decoding module (Decoder) part, and the structure of the inter noise reduction self-Encoder is shown in figure 2.
The noise reduction self-Encoder is a self-supervision learning model which trains the original data as a label, encodes the original input added with noise into hidden layer output in an Encoder, and reconstructs the hidden output into the original input in the Decoder, so that the dimensional compression and enhancement of the characteristics can be realized. The hidden layer output in the noise reduction self-encoder is the final characteristic after enhancement.
D. Fused attention time convolution network
The design of a fusion attention time convolution network structure;
in order to accurately grasp the conditions of equipment state, silk drying process, environmental influence and the like, in the embodiment, a fused attention time convolution network is designed to realize mode discrimination based on monitoring data, and the adjustment time and adjustment quantity of each parameter are intelligently identified.
Specifically, the fused attention time convolution network comprises a self-attention part and a time convolution part, and the real-time monitoring data is assumed to be transformed into the data after cross-derivatization and characteristic enhancementWherein N is the number of features, and T is the time span.
In the self-attention part, three transformation matrices w are defined q ,w k ,w v And performing linear transformation on X for three times to obtain a derivative matrix of X, and respectively marking the derivative matrix as a query matrix Q, a key matrix K and a value matrix V. Then, updating the value matrix by using the query matrix and the key matrix, so that the model selectively focuses on information of different dimensions and different moments, and simultaneously, corresponding weights are distributed for each position of the input X, and the self-focusing calculation result is as follows:
R=σ[QK T ]V;
where σ is a nonlinear activation function, the value range can be mapped between (0, 1).
In this embodiment, σ is a Sigmoid function:
the Sigmoid function is most sensitive at the 0.5 position, and can well distinguish the part needing to be reserved or inhibited in the value matrix.
In the time convolution part, convolution operation is often used for processing an image, in this embodiment, the size of a convolution kernel in a feature dimension is set to 1, and the size of a time dimension is set to t, so that the convolution operation is performed only in the time dimension, and time-varying information of the feature can be effectively extracted. The process of drying the shreds is continuous in time, and time-varying information of mastering data is very important for adjusting control parameters.
Specifically, the convolution kernel is:
U=[u 1 ,u 2 ,…,u t ];
the self-attention calculation result is convolved to obtain:
Y=R⊙U;
wherein, as follows, the convolution operation is performed.
Finally, the output Y of the time convolution is transmitted into a fully-connected neural network, and N characteristics are fused and transformed into one-dimensional data:
F=w f Y+b;
wherein w is f And b is an offset, which is the weight of the fully connected neural network. F is the final output of the fused attention time convolution network.
E. Fused attention time convolutional network training
E1. Constructing training samples and labels;
the history monitoring data are subjected to the preprocessing and the characteristic engineering, the history monitoring data are divided according to time intervals of 120s to be used as training samples, the adjustment quantity of each control parameter is calculated at the 120s of each sample to be used as a label, and four control parameters including material flow, drying temperature, combustion gas flow and steam flow are arranged in the process of drying the silk, namely, one sample corresponds to four labels.
E2. Training a network;
then training the fused attention time convolution network according to different labels, in the embodiment, updating model parameters by adopting an Adam optimizer, wherein the updating process is as follows:
m t =δ 1 ·m t-1 +(1-δ 1 )·g t ;
wherein g t For the gradient of the objective function at time t, m t For first order matrix estimation, n t Delta for second order matrix estimation 1 And delta 2 Exponential decay Rate, gamma, estimated for matrix t For the parameters to be updated, α is the learning rate and ε is a small constant added to maintain data stability.
And carrying out iterative updating on the model parameters through an Adam optimizer to complete training of the model, and obtaining a model discrimination model corresponding to four control parameters of material flow, drying temperature, combustion gas temperature and steam flow. The real-time monitoring data of the current batch is input into the model, so that the intelligent adjustment of the control parameters can be realized, and the aim of controlling the water content of the outlet is fulfilled.
On the basis of data preprocessing and characteristic engineering, the invention designs a fused attention time convolution network to realize mode discrimination of the silk drying process, and intelligently adjusts control parameters so as to realize control of outlet water content.
The control method disclosed by the invention can accurately grasp the equipment state, the drying process and the environmental influence, and timely adjust the control parameters, compared with PID control and manual adjustment, the control method has the advantages of no time lag, higher adjustment rationality and stronger stability and accuracy of the outlet water content. Fig. 3-4 are graphs comparing the outlet water content adjusted by the method with the outlet water content adjusted by PID/manual adjustment, wherein the multi-section broken line is the actual value of the outlet water content, and the horizontal line is the standard value of the outlet water content, so that it is obvious that the stability of the outlet water content is stronger and the accuracy is higher.
In order to quantify the superiority of the method compared with the PID/manual adjustment method, 5 batch production experiments are respectively carried out under the actual production scene, the variance of the outlet water content of each batch and the average deviation of the actual value and the standard value of the outlet water content are calculated, and the experimental results are shown in table 1.
Table 1 experimental results
From Table 1, the variance and average deviation of the outlet water content under the adjustment of the method are obviously smaller than those of PID/manual adjustment, which shows that the method can obviously improve the accuracy and stability of the outlet water content of the dried shreds.
In the case of example 2,
the overall process from data acquisition to model output control parameter adjustment scheme is shown in fig. 5, and the method in embodiment 1 is realized by a drying wire outlet water content control system based on mode discrimination, which comprises a data acquisition and storage module, a data preprocessing module, a characteristic cross derivatization and enhancement module, a fused attention time convolution network module and a fused attention time convolution network training module, wherein the data acquisition and storage module is used for monitoring data acquisition and data storage in real time, the data preprocessing module is used for abnormal moment identification, missing value filling and data normalization processing, the characteristic cross derivatization and enhancement module is used for characteristic cross derivatization and characteristic enhancement, the fused attention time convolution network module is used for designing a fused attention time convolution network structure, and the fused attention time convolution network training module is used for constructing training samples and labels and training a fused attention time convolution network.
Although the invention has been described herein with reference to the above-described illustrative embodiments thereof, the above-described embodiments are merely preferred embodiments of the present invention, and the embodiments of the present invention are not limited by the above-described embodiments, it should be understood that numerous other modifications and embodiments can be devised by those skilled in the art that will fall within the scope and spirit of the principles of this disclosure.
Claims (10)
1. The method for controlling the water content of the outlet of the baked silk based on mode discrimination is characterized by comprising the following specific steps:
A. data acquisition and storage
The method comprises the steps of monitoring data acquisition in real time, wherein the acquired data comprises monitoring data and control parameters;
recording the specific values of the monitoring data and the control parameters of each batch according to the production plan, forming historical monitoring data of the finished batch, and forming real-time monitoring data of the current production batch;
storing data, namely storing monitoring data by using a MySQL database;
B. data preprocessing
Abnormal moment identification, namely eliminating abnormal values causing data abnormality by locating the abnormal moment in the data acquisition process;
filling the missing value, filling the missing value and the deleted monitoring data in abnormal moment identification in the data acquisition process, training a random forest regression model by taking the complete historical monitoring data as a training set, and predicting and filling the missing value and the deleted detection data;
normalizing the data, and normalizing the monitoring data;
C. feature cross-derivatization and enhancement
Feature cross derivatization, namely performing feature cross derivatization on the monitoring data after data preprocessing, mining internal connection among the monitoring data, specifically encoding a nonlinear rule in a feature space through multi-feature Cartesian product, and generating a synthesized feature, thereby increasing the dimension of the feature;
feature enhancement, namely performing degradation and enhancement on derived features, specifically designing a noise reduction self-encoder for the characteristics of the monitored data, and performing depth fusion and enhancement on the cross derived features through the noise reduction self-encoder;
D. fused attention time convolution network
Designing a structure of a convolution network integrating attention time, realizing mode discrimination based on monitoring data, and intelligently identifying adjustment time and adjustment quantity of each parameter;
E. fused attention time convolutional network training
Constructing training samples and labels, dividing historical monitoring data into training samples according to a certain time after preprocessing, feature cross derivatization and enhancement, and calculating the adjustment quantity of each control parameter at the end of a certain time of each training sample to serve as the labels, wherein one training sample corresponds to a plurality of labels;
then training the fused attention time convolution network according to different labels respectively to obtain mode discrimination models corresponding to a plurality of labels;
the real-time monitoring data of the current batch is input into a mode discrimination model, so that the intelligent adjustment of control parameters is realized, and the control of the water content of the outlet is realized.
2. The method for controlling the water content of the outlet of the cut tobacco dryer based on mode discrimination according to claim 1, wherein the monitoring data at least comprises the water content of the outlet, the outlet temperature, the water content of the inlet, the inlet temperature, the return gas temperature, the air temperature, the negative moisture removal pressure, the ambient temperature and the ambient humidity; the control parameters at least comprise material flow, drying temperature, combustion gas flow and steam flow; the total of the monitoring data and the control parameters is not less than 13, that is to say the dimension of the monitoring data is at least 13.
3. The method for controlling the water content of the outlet of the baked wire based on the mode discrimination according to claim 1, wherein the MySQL database divides the collected data according to the date, and stores the collected data of different dates into different tables.
4. The method for controlling the moisture content of the outlet of the baked yarn based on the mode discrimination according to claim 3, wherein the monitoring data is stored by using a MySQL database, and the method comprises the following specific steps: and acquiring and writing the monitoring data every second, forming monitoring data with the time interval of 1s in a MySQL database, assigning unique ids for each batch for distinguishing, and simultaneously creating a group_record table to record all batch information, wherein the group_record table at least comprises batch id, start time, end time and material license plate information.
5. The method for controlling the water content of the outlet of the baked wire based on the mode discrimination according to claim 1, wherein the abnormal time is identified, and the abnormal value causing the data abnormality is removed by locating the abnormal time in the data acquisition process;
the abnormal moment is positioned by means of multi-element Gaussian distribution detection, and the method specifically comprises the following steps:
assume that the collected monitoring data set is:
wherein n is the dimension of the monitoring data;
the mean vector of n dimensions is calculated as:
wherein E (x) i ) The average value of the i-th dimension monitoring data is m, and the m is the number of certain monitoring data;
thereby calculating the monitoring dataIs a covariance matrix of (a):
then the data is monitoredThe distribution probability of (2) is:
probability of distribution of the monitored data for each instant, ifThe time is marked as an abnormal time, and the monitoring data at the time is deleted.
6. The method for controlling the water content of the outlet of the baked wire based on the mode discrimination according to claim 1, wherein the data normalization is performed on the monitored data by using a maximum and minimum value, and the method is characterized by using the following formula:
wherein x is * For normalized monitoring data, x is the original monitoring data, min (x) is the minimum value of the original monitoring data, and max (x) is the maximum value of the original monitoring data.
7. The method for controlling the water content of the outlet of the baked wire based on the mode discrimination according to claim 1, wherein the noise reduction self-encoder consists of three parts, namely a random noise introduction part, an encoding module and a decoding module; the noise reduction self-Encoder is a self-supervision learning model which trains the original data as a label, encodes the original input added with noise into hidden layer output in an Encoder, and reconstructs the hidden output into the original input in the Decoder so as to realize the dimensional compression and enhancement of the characteristics; the hidden layer output in the noise reduction self-encoder is the final characteristic after enhancement.
8. The method for controlling the water content of a cut tobacco outlet based on mode discrimination according to claim 1, wherein the fused attention time convolution network comprises a self-attention part and a time convolution part, and real-time monitoring data is assumed to be transformed into the data after cross derivatization and characteristic enhancementWherein N is the number of features, and T is the time span;
in the self-attention part, three transformation matrices w are defined q ,w k ,w v Performing linear transformation on X for three times to obtain a derivative matrix of X, and respectively marking the derivative matrix as a query matrix Q, a key matrix K and a value matrix V; then, updating the value matrix by using the query matrix and the key matrix, so that the model selectively focuses on information of different dimensions and different moments, and simultaneously, corresponding weights are distributed for each position of the input X, and the self-focusing calculation result is as follows:
R=σ[QK T ]V;
wherein σ is a nonlinear activation function that maps the value range between (0, 1);
sigma adopts a Sigmoid function:
the Sigmoid function is most sensitive at the position of 0.5, and can well distinguish the part needing to be reserved or inhibited in the value matrix;
in the time convolution part, convolution operation is commonly used for processing an image, the size of a convolution kernel in a characteristic dimension is set to be 1, and the size of a time dimension is set to be t, so that the convolution operation is only performed in the time dimension;
specifically, the convolution kernel is:
U=[u 1 ,u 2 ,…,u t ];
the self-attention calculation result is convolved to obtain:
Y=R⊙U;
wherein, as follows;
finally, the output Y of the time convolution is transmitted into a fully-connected neural network, and N characteristics are fused and transformed into one-dimensional data:
F=w f Y+b;
wherein w is f And b is an offset, and F is the final output of the fusion attention time convolution network.
9. The method for controlling the water content of the outlet of the baked wire based on the mode discrimination according to claim 1, wherein the network trains, and then trains the fused attention time convolution network according to different labels respectively to obtain a mode discrimination model corresponding to a plurality of labels;
updating model parameters by adopting an Adam optimizer to finish training the model, wherein the updating process is as follows:
m t =δ 1 ·m t-1 +(1-δ 1 )·g t ;
wherein g t For the gradient of the objective function at time t, m t For first order matrix estimation, n t Delta for second order matrix estimation 1 And delta 2 Exponential decay Rate, gamma, estimated for matrix t For the parameters to be updated, α is the learning rate and ε is a small constant added to maintain data stability.
10. The control system for the water content of the outlet of the drying wire based on the mode discrimination is used for realizing the control method for the water content of the outlet of the drying wire based on the mode discrimination according to any one of claims 1-9 and is characterized by comprising a data acquisition and storage module, a data preprocessing module, a characteristic cross derivatization and enhancement module, a fused attention time convolution network module and a fused attention time convolution network training module, wherein the data acquisition and storage module is used for monitoring data acquisition and data storage in real time, the data preprocessing module is used for abnormal moment identification, missing value filling and data normalization processing, the characteristic cross derivatization and enhancement module is used for characteristic cross derivatization and characteristic enhancement, the fused attention time convolution network module is used for designing a fused attention time convolution network structure, and the fused attention time convolution network training module is used for constructing training samples and labels and training a fused attention time convolution network.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311193095.9A CN117289594A (en) | 2023-09-15 | 2023-09-15 | Method and system for controlling water content of outlet of cut tobacco drying based on mode discrimination |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311193095.9A CN117289594A (en) | 2023-09-15 | 2023-09-15 | Method and system for controlling water content of outlet of cut tobacco drying based on mode discrimination |
Publications (1)
Publication Number | Publication Date |
---|---|
CN117289594A true CN117289594A (en) | 2023-12-26 |
Family
ID=89247303
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311193095.9A Pending CN117289594A (en) | 2023-09-15 | 2023-09-15 | Method and system for controlling water content of outlet of cut tobacco drying based on mode discrimination |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117289594A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117949976A (en) * | 2024-03-26 | 2024-04-30 | 山西省交通运输运行监测与应急处置中心(山西省交通运输厅新闻中心) | Real-time positioning method and system for highway transport vehicle for transporting yellow phosphorus |
-
2023
- 2023-09-15 CN CN202311193095.9A patent/CN117289594A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117949976A (en) * | 2024-03-26 | 2024-04-30 | 山西省交通运输运行监测与应急处置中心(山西省交通运输厅新闻中心) | Real-time positioning method and system for highway transport vehicle for transporting yellow phosphorus |
CN117949976B (en) * | 2024-03-26 | 2024-06-11 | 山西省交通运输运行监测与应急处置中心(山西省交通运输厅新闻中心) | Real-time positioning method and system for highway transport vehicle for transporting yellow phosphorus |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108021947B (en) | A kind of layering extreme learning machine target identification method of view-based access control model | |
CN117289594A (en) | Method and system for controlling water content of outlet of cut tobacco drying based on mode discrimination | |
CN115494865B (en) | Unmanned aerial vehicle cluster situation analysis method and medium based on space-time diagram convolutional network | |
CN107622236B (en) | Crop disease diagnosis and early warning method based on swarm and gradient lifting decision tree algorithm | |
CN110782658A (en) | Traffic prediction method based on LightGBM algorithm | |
CN111237988A (en) | Control method and system for subway vehicle-mounted air conditioning unit | |
CN114548591B (en) | Sequential data prediction method and system based on mixed deep learning model and Stacking | |
CN117073768B (en) | Beef cattle cultivation management system and method thereof | |
CN114015825A (en) | Method for monitoring abnormal state of blast furnace heat load based on attention mechanism | |
CN115526298A (en) | High-robustness comprehensive prediction method for concentration of atmospheric pollutants | |
CN114403486B (en) | Intelligent control method of airflow type cut-tobacco drier based on local peak value coding circulation network | |
CN116484289A (en) | Carbon emission abnormal data detection method, terminal and storage medium | |
CN116340746A (en) | Feature selection method based on random forest improvement | |
CN113984989A (en) | Aquaculture water quality abnormity detection method based on Laplace dimensionality reduction | |
CN116649191A (en) | Remote fertilization and irrigation control system and method based on PLC | |
CN116414095A (en) | Data-driven optimization method for technological parameters in traditional Chinese medicine manufacturing process | |
CN114118401A (en) | Neural network-based power distribution network flow prediction method, system, device and storage medium | |
CN113177578A (en) | Agricultural product quality classification method based on LSTM | |
CN111838744A (en) | Continuous real-time prediction method for moisture of environment temperature and humidity in tobacco shred process based on LSTM | |
CN117010263A (en) | Residual life prediction method based on convolutional neural network and long-term and short-term memory network | |
CN114707931A (en) | HTM-based real-time monitoring system for abnormal conditions of logistics warehouse and design method thereof | |
CN117370870B (en) | Knowledge and data compound driven equipment multi-working condition identification and performance prediction method | |
CN112508320B (en) | Automatic process stage division workflow for batch production | |
CN118041689B (en) | Network malicious traffic detection method | |
CN114580295B (en) | Pressure working condition identification method based on elastic BP random forest fusion |
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 |