CN103267826A - Soft measurement method for online detection of gelatin concentration - Google Patents
Soft measurement method for online detection of gelatin concentration Download PDFInfo
- Publication number
- CN103267826A CN103267826A CN2013101317090A CN201310131709A CN103267826A CN 103267826 A CN103267826 A CN 103267826A CN 2013101317090 A CN2013101317090 A CN 2013101317090A CN 201310131709 A CN201310131709 A CN 201310131709A CN 103267826 A CN103267826 A CN 103267826A
- Authority
- CN
- China
- Prior art keywords
- data
- neural network
- gelatin
- model
- temperature
- 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
Images
Landscapes
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
A soft measurement method for online detection of a gelatin concentration relates to a method of detecting the gelatin concentration. The method comprises acquisition of process data, pretreatment of sample data, establishment of soft measurement model, and calculation of the gelatin solution concentration by using the model, wherein the acquisition comprises acquiring the temperature, time, and density values, which are used as auxiliary variables of modeling of the soft measurement, of the gelatin solution on site under different conditions; the pretreatment comprises treatment of abnormity data, treatment of random errors and treatment of normalization; the establishment comprises establishing an Elman neural network model topological structure, determining each single neural network model structure and training algorithm; and the calculation comprises introducing obtained weight coefficients of each single neural network model into a combined neural network model, calculating corresponding values by using the combined model, and performing an anti-normalization treatment for the values, thereby obtaining the gelatin concentration. The method realizes online real-time automatic detection for the gelatin concentration, and provides detection data for realizing automatic control during a gelatin production process.
Description
Technical field
The present invention relates to a kind of method that detects gelatin concentration, particularly relate to a kind of flexible measurement method of online detection gelatin concentration.
Background technology
Gelatin enterprise is in producing the gelatin process, and glue-extracting working procedure is center operation crucial in producing, and the speed of glue-extracting working procedure progress can have influence on whole manufacturing of gelatin speed.And gelatin concentration is to carry most important technological parameter in the glue process, and it directly affects the quality of production of gelatin, has very important directive significance in manufacturing of gelatin.
At present, gelatin enterprise adopts the method for artificial timing sampling, off-line measurement to the mensuration of gelatin concentration, but because sampling time interval is longer, and labor measurement is bigger time lag, can not provide the detection data for controlling automatically, to instructing production unfavorable.Therefore, realize the online real-time detection to gelatin concentration, most important to gelatin enterprise improve production efficiency and product quality.And flexible measurement method is realize at present that gelatin concentration measures the most effective, most economical, detection method the most efficiently.
Summary of the invention
The object of the present invention is to provide a kind of flexible measurement method of online detection gelatin concentration.This method has realized the online real time automatic detection of gelatin concentration, realizes that for gelatine production course control provides the detection data automatically, can in time instruct manufacturing of gelatin, is conducive to improve the manufacturing of gelatin quality and enhances productivity.
The objective of the invention is to be achieved through the following technical solutions:
A kind of flexible measurement method of online detection gelatin concentration said method comprising the steps of: carry out the collection of process data, soft-sensing model is set up in the sampled data pre-service, utilizes model to calculate gelatin solution concentration; The collection of carrying out process data is: gelatin solution temperature, time and density value are as the auxiliary variable of soft sensor modeling under the different operating modes of collection in worksite; Sampled data pre-service: the processing, the processing of stochastic error, the normalized that comprise abnormal data; Setting up soft-sensing model is: set up combination Elman neural network model topological structure, determine each single neural network model structure and training algorithm, carry out determining of combination neural net weights; Utilize model to calculate gelatin solution concentration to be: bring the weighting coefficient of each single neural network model of trying to achieve into the combination neural net model, when built-up pattern is input as gelatin solution density, temperature and corresponding sampling time, utilize this built-up pattern can calculate respective value, after this numerical value carried out anti-normalized, just obtain the gelatin concentration value.
The flexible measurement method of described a kind of online detection gelatin concentration, in the described collection of carrying out process data, temperature reads by temperature sensor, detects the concentration value of gelatin solution by the hand-held spectroscope, and measures the density of solution with densitometer.
The flexible measurement method of described a kind of online detection gelatin concentration, the processing of described abnormal data are according to the standard deviation of Bessel formula accounting temperature (T):
If a certain temperature samples data
Deviation
(
) satisfy:
, then think
Be abnormal data, rejected.
The flexible measurement method of described a kind of online detection gelatin concentration, the processing of described stochastic error will be adopted the average value filtering algorithm respectively through remaining temperature, density and viscosity data after the Error processing, remove the stochastic error in the sampled data.
The flexible measurement method of described a kind of online detection gelatin concentration, described normalized will be passed through
R temperature data after stochastic error is handled
Normalize in [0,1] interval and be:
Wherein,
Be
It is individual through the temperature data after the normalized,
Be the minimum value of r temperature sampling data,
Maximal value for r sampled data.
The flexible measurement method of described a kind of online detection gelatin concentration, described foundation combination Elman neural network model topological structure is combined a plurality of Elman neural network models, constitutes the combination neural net model.
The flexible measurement method of described a kind of online detection gelatin concentration, described definite each single neural network model structure and training algorithm, r after a normalized sample data is divided into two groups: one group of common initial training data set as each single neural network model, respectively each single Elman neural network is trained; Another group is used for the neural network model that verification is set up as the checking data collection.
The flexible measurement method of described a kind of online detection gelatin concentration, its combination neural net weights determine that each Determination of Weight Coefficient adopts and minimizes maximum absolute error as finding the solution the standard of the combining weights of combination neural net.
Further specify as follows:
1. the collection of process data
Gelatin solution temperature (T), time (t) and density under the different operating modes of collection in worksite (
) value is as the auxiliary variable of soft sensor modeling.Wherein, temperature (T) can read by temperature sensor, and records the corresponding time (t), simultaneously, detects the concentration value (c) of gelatin solution by the hand-held spectroscope, and with densitometer measure solution density (
).With gelatin solution temperature, time, density and the concentration value of synchronization record as one group of sample data (T, t,
, c), every about half an hour, repeat this operation, up to collect needed sample number till.
2. sampled data pre-service
Respectively organize temperature value to what collect
, density value
And viscosity number
Calculating its arithmetic mean value respectively is
,
With
And calculate its deviate respectively
,
,
(
), according to the standard deviation of Bessel formula accounting temperature (T):
If a certain temperature samples data
Deviation
(
) satisfy:
, then think
Be abnormal data, rejected.In like manner, calculate the standard deviation of density, viscosity sample data respectively
With
, rejecting abnormal data wherein uses the same method.
To pass through
Remaining r group temperature (T) after the Error processing, density (
) and viscosity (c) data adopt the average value filtering algorithm respectively, to remove the stochastic error in the sampled data, improve the used data precision of following model.
To pass through
R temperature data after stochastic error is handled
Normalize in [0,1] interval and be:
Wherein,
Be
It is individual through the temperature data after the normalized,
Be the minimum value of r temperature sampling data,
Maximal value for r sampled data.Use the same method, r density, viscosity data and a corresponding r time data are normalized to respectively in [0,1] interval, obtain density, viscosity and time data after the normalization
,
,
(
).
3. set up soft-sensing model
The present invention combines a plurality of Elman neural network models, constitutes the combination neural net model,
See shown in Figure 1.Total output of combination neural net model is the weighted sum of each neural network output, that is:
Wherein X is the input data matrix of neural network,
Be combination neural net output,
Be the forecast model of combination neural net,
Be the Elman network number of combination,
Be
Individual neural network prediction model,
Be
The non-negative weighting coefficient of individual Elman neural network satisfies normalizing, that is:
The topological structure of each single Elman neural network model as shown in Figure 2, its mathematical model is:
Wherein
Connection weight matrix for context unit and hidden layer unit;
Connection weight matrix for input block and hidden layer unit;
Connection weight matrix for hidden layer unit and output unit;
The input of expression Elman network is 3 dimensional vectors;
With
Be respectively the output of n dimension context unit and hidden layer unit;
Being the output of output layer unit, is 1 dimensional vector;
Be the neuronic transfer function of output layer;
Transfer function for hidden layer neuron.
The Elman network carries out the correction of weights by the gradient descent method, and the study index adopts sum of squared errors function to represent:
The learning algorithm of Elman network is:
R after a normalized sample data is divided into two groups: one group of common initial training data set as each single neural network model, respectively each single Elman neural network is trained; Another group is used for the neural network model that verification is set up as the checking data collection.
, the determining of combination neural net weights
Total output of combination neural net
Be the weighted sum of each single Elman neural network model output, the present invention adopts for each Determination of Weight Coefficient and minimizes maximum absolute error as finding the solution the standard of the combining weights of combination neural net.Concrete solution procedure is as follows:
If
Be of combination neural net forecast model
The combination neural net error of individual data point, then:
In the formula
mFor checking data is concentrated the sample data number;
Expression the
iIndividual Elman neural network is
Individual data point error;
With
All represent
The sample output valve of individual data point;
Expression the
The combination neural net model predication value of individual data point.If
The maximal value of the predicated error absolute value of expression combination neural net model, then
Reach minimum linear combination forecasting model with the maximum error absolute value and can be expressed as following optimization problem:
Find the solution this optimization problem and can obtain the weighting coefficient of each single neural network.
4. utilize model to calculate gelatin solution concentration
Bring the weighting coefficient of each single neural network model of trying to achieve into the combination neural net model, when built-up pattern is input as gelatin solution density, temperature and corresponding sampling time, utilize this built-up pattern can calculate respective value, after this numerical value carried out anti-normalized, just can obtain the gelatin concentration value.
Description of drawings
Fig. 1 is each single neural network model topology diagram;
Fig. 2 is combination neural net model topology structural drawing.
Embodiment
Come below in conjunction with specific embodiments content of the present invention is further described.
(1) collection of sample data
With gelatin solution density (
), temperature (T) and corresponding sampling time (
t) be input variable,
Gelatin solution concentration (c) is output variable.Temperature data passes through the temperature sensor collection in worksite, and records the corresponding time; Simultaneously, glue is sampled, detect by the hand-held spectroscope and obtain concentration data, and measure the density originally of sampling with densitometer.With the glue temperature of synchronization record, time, and the gelatin concentration of manual detection and density as one group of sample data, every about half an hour, repeat this operation, the 235 groups of sample datas of sampling altogether.
(2) sample Data Preprocessing
To 235 groups of temperature values that collect
, density value
And viscosity number
Calculating its arithmetic mean value respectively is
,
, and
And calculate its deviate respectively
,
,
(
), according to the standard deviation of Bessel formula accounting temperature (T):
If a certain temperature samples data
Deviation
(
) satisfy:
, then think
Be abnormal data, rejected.In like manner, calculate the standard deviation of density, viscosity sample data respectively
With
, rejecting abnormal data wherein uses the same method.
235 groups of data are carried out abnormal data reject 220 groups of remaining sampled datas utilization average value filtering algorithms of back, to eliminate the stochastic error that exists in the data.
220 groups of data after handling through average value filtering are carried out normalized: with 220 temperature datas
Normalize in [0,1] interval and be:
Wherein,
Be the minimum value of 220 temperature sampling data,
Be the maximal value of 220 sampled datas,
For through the after the normalized
Individual temperature data.Use the same method, 220 density datas, 220 carried the glue time data and 220 viscosity datas normalize to respectively in [0,1] interval, obtain after the normalization density, carry glue time and viscosity data
,
With
,
Temperature data with the synchronization after the normalized
, density data
,
Data and correspondingly carry the glue time
Sample data of data composition (
,
,
,
) the final sample data collection of formation.
Concentrating arbitrarily from sample data, 170 groups of sample datas of extraction are used for neural network training to set up soft-sensing model as learning sample; Remaining 50 groups as the generalization ability of verification sample with the check soft-sensing model.
(3) combination neural net structure of models
5 single Elman neural network models are combined, constituted the combination neural net model.The input of combination neural net is identical with the input data of each single neural network model, and total output of combination neural net is the weighted sum of various neural network outputs, that is:
(4) each single Elman neural network model structure
5 single Elman neural network models adopt 3 * 5 * 1 respectively, 3 * 6 * 1,3 * 7 * 1,3 * 8 * 1,3 * 9 * 1 structure, be that each single Elman neural network is divided into 3 layers: input layer is 3 neurons, and output layer is 1 neuron, and hidden layer adopts 5,6,7,8 and 9 neurons respectively.Hidden layer adopts
The type transfer function, output layer adopts
The type transfer function.
(5) training of single Elman neural network model
Respectively 5 single Elman neural networks are trained with 170 groups of training datas, the termination error of training is 1 * 10
-3, by training the weight of determining each single Elman neural network.
(6) combination neural net model combination coefficient determines
(7) calculate gelatin solution concentration
With flexible measurement method of the present invention, be used for detecting in real time the gelatin solution concentration value, and with itself and hand-held
Spectroscope detects and obtains concentration data and compare, and partial results is as shown in table 1.The gelatin solution concentration value that calculates with flexible measurement method of the present invention and the detected value of hand-held spectroscope are very approaching, and error all within ± 2.3%, satisfies demand of practical production fully, and production is had actual directive significance.
Table 1 combination neural net part predicted value and measured value comparison sheet
Claims (8)
1. the flexible measurement method of an online detection gelatin concentration is characterized in that, said method comprising the steps of: carry out the collection of process data, soft-sensing model is set up in the sampled data pre-service, utilizes model to calculate gelatin solution concentration; The collection of carrying out process data is: gelatin solution temperature, time and density value are as the auxiliary variable of soft sensor modeling under the different operating modes of collection in worksite; Sampled data pre-service: the processing, the processing of stochastic error, the normalized that comprise abnormal data; Setting up soft-sensing model is: set up combination Elman neural network model topological structure, determine each single neural network model structure and training algorithm, carry out determining of combination neural net weights; Utilize model to calculate gelatin solution concentration to be: bring the weighting coefficient of each single neural network model of trying to achieve into the combination neural net model, when built-up pattern is input as gelatin solution density, temperature and corresponding sampling time, utilize this built-up pattern can calculate respective value, after this numerical value carried out anti-normalized, just obtain the gelatin concentration value.
2. the flexible measurement method of a kind of online detection gelatin concentration according to claim 1, it is characterized in that in the described collection of carrying out process data, temperature reads by temperature sensor, detect the concentration value of gelatin solution by the hand-held spectroscope, and measure the density of solution with densitometer.
3. the flexible measurement method of a kind of online detection gelatin concentration according to claim 1 is characterized in that, the processing of described abnormal data is according to the standard deviation of Bessel formula accounting temperature (T):
If a certain temperature samples data
Deviation
(
) satisfy:
, then think
Be abnormal data, rejected.
4. the flexible measurement method of a kind of online detection gelatin concentration according to claim 1, it is characterized in that, the processing of described stochastic error will be adopted the average value filtering algorithm respectively through remaining temperature, density and viscosity data after the Error processing, remove the stochastic error in the sampled data.
5. the flexible measurement method of a kind of online detection gelatin concentration according to claim 1 is characterized in that, described normalized will be passed through
R temperature data after stochastic error is handled
Normalize in [0,1] interval and be:
6. the flexible measurement method of a kind of online detection gelatin concentration according to claim 1 is characterized in that, described foundation combination Elman neural network model topological structure is combined a plurality of Elman neural network models, constitutes the combination neural net model.
7. the flexible measurement method of a kind of online detection gelatin concentration according to claim 1, it is characterized in that, described definite each single neural network model structure and training algorithm, r after a normalized sample data is divided into two groups: one group of common initial training data set as each single neural network model, respectively each single Elman neural network is trained; Another group is used for the neural network model that verification is set up as the checking data collection.
8. the flexible measurement method of a kind of online detection gelatin concentration according to claim 1, it is characterized in that, determining of described combination neural net weights, each Determination of Weight Coefficient adopt and minimize maximum absolute error as finding the solution the standard of the combining weights of combination neural net.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2013101317090A CN103267826A (en) | 2013-04-17 | 2013-04-17 | Soft measurement method for online detection of gelatin concentration |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2013101317090A CN103267826A (en) | 2013-04-17 | 2013-04-17 | Soft measurement method for online detection of gelatin concentration |
Publications (1)
Publication Number | Publication Date |
---|---|
CN103267826A true CN103267826A (en) | 2013-08-28 |
Family
ID=49011465
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2013101317090A Pending CN103267826A (en) | 2013-04-17 | 2013-04-17 | Soft measurement method for online detection of gelatin concentration |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103267826A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103471971A (en) * | 2013-09-26 | 2013-12-25 | 沈阳大学 | Soft measurement method for online detecting fine powder rate of aluminum powder |
CN108983849A (en) * | 2018-07-12 | 2018-12-11 | 沈阳大学 | It is a kind of to utilize compound extreme learning machine ANN Control greenhouse method |
CN110322933A (en) * | 2019-06-20 | 2019-10-11 | 浙江工业大学 | A kind of polypropylene melt index hybrid modeling method based on dynamic error compensation mechanism |
CN111914476A (en) * | 2020-06-23 | 2020-11-10 | 宁波大学 | Online soft measurement method for butane content of product at bottom of debutanizer |
CN111912875A (en) * | 2020-06-23 | 2020-11-10 | 宁波大学 | Fractionating tower benzene content soft measurement method based on stack type Elman neural network |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH07129532A (en) * | 1993-11-08 | 1995-05-19 | Meidensha Corp | Neural network |
EP2143049A1 (en) * | 2007-04-27 | 2010-01-13 | Siemens Aktiengesellschaft | A method for computer-assisted learning of one or more neural networks |
EP2146641A1 (en) * | 2007-04-30 | 2010-01-27 | Uniwersytet Im. Kazimierza Wielkiego W Bydgoszczy | A method for the non-invasive determination of density and poroelasticity parameters of the long bone |
CN101718769A (en) * | 2009-11-17 | 2010-06-02 | 重庆大学 | Parallel type genetic Elman neural network-based source driving 235U concentration recognition method |
CN202472729U (en) * | 2011-12-20 | 2012-10-03 | 西北师范大学 | SIMF Elman network structure |
CN102937670A (en) * | 2012-11-17 | 2013-02-20 | 河南科技大学 | Method for identifying resistance parameters of rotors of induction motor on basis of Elman neural network |
-
2013
- 2013-04-17 CN CN2013101317090A patent/CN103267826A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH07129532A (en) * | 1993-11-08 | 1995-05-19 | Meidensha Corp | Neural network |
EP2143049A1 (en) * | 2007-04-27 | 2010-01-13 | Siemens Aktiengesellschaft | A method for computer-assisted learning of one or more neural networks |
EP2146641A1 (en) * | 2007-04-30 | 2010-01-27 | Uniwersytet Im. Kazimierza Wielkiego W Bydgoszczy | A method for the non-invasive determination of density and poroelasticity parameters of the long bone |
CN101718769A (en) * | 2009-11-17 | 2010-06-02 | 重庆大学 | Parallel type genetic Elman neural network-based source driving 235U concentration recognition method |
CN202472729U (en) * | 2011-12-20 | 2012-10-03 | 西北师范大学 | SIMF Elman network structure |
CN102937670A (en) * | 2012-11-17 | 2013-02-20 | 河南科技大学 | Method for identifying resistance parameters of rotors of induction motor on basis of Elman neural network |
Non-Patent Citations (2)
Title |
---|
曹洁等: "基于RBFNN模型明胶浓度在线测量", 《电气自动化》, vol. 34, no. 5, 30 September 2012 (2012-09-30) * |
高利敏: "明胶浓度的软测量建模及参数优化", 《中国优秀硕士学位论文全文数据库(信息科技辑)》, no. 9, 15 September 2011 (2011-09-15) * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103471971A (en) * | 2013-09-26 | 2013-12-25 | 沈阳大学 | Soft measurement method for online detecting fine powder rate of aluminum powder |
CN108983849A (en) * | 2018-07-12 | 2018-12-11 | 沈阳大学 | It is a kind of to utilize compound extreme learning machine ANN Control greenhouse method |
CN110322933A (en) * | 2019-06-20 | 2019-10-11 | 浙江工业大学 | A kind of polypropylene melt index hybrid modeling method based on dynamic error compensation mechanism |
CN111914476A (en) * | 2020-06-23 | 2020-11-10 | 宁波大学 | Online soft measurement method for butane content of product at bottom of debutanizer |
CN111912875A (en) * | 2020-06-23 | 2020-11-10 | 宁波大学 | Fractionating tower benzene content soft measurement method based on stack type Elman neural network |
CN111914476B (en) * | 2020-06-23 | 2023-09-29 | 宁波大学 | On-line soft measurement method for butane content of debutanizer bottom product |
CN111912875B (en) * | 2020-06-23 | 2024-02-13 | 江苏淮河化工有限公司 | Fractionation tower benzene content soft measurement method based on stacked Elman neural network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN114488988B (en) | Industrial Internet of things for production line balance rate regulation and control method | |
CN108898215B (en) | Intelligent sludge bulking identification method based on two-type fuzzy neural network | |
CN103267826A (en) | Soft measurement method for online detection of gelatin concentration | |
CN103745273B (en) | Semiconductor fabrication process multi-performance prediction method | |
CN110807554B (en) | Generation method and system based on wind power/photovoltaic classical scene set | |
CN105956702A (en) | Aquaculture water quality dissolved oxygen prediction method based on data fusion | |
CN102183621A (en) | Aquaculture dissolved oxygen concentration online forecasting method and system | |
CN107169610A (en) | Aquaculture dissolved oxygen prediction method and device | |
CN110751318A (en) | IPSO-LSTM-based ultra-short-term power load prediction method | |
CN110322014A (en) | A kind of finished cement specific surface area prediction technique based on BP neural network | |
CN109389238B (en) | Ridge regression-based short-term load prediction method and device | |
CN112836884A (en) | Copula-DBiLSTM-based method for accurately predicting multi-element load of comprehensive energy system | |
CN104865827B (en) | A kind of pumping production optimization method based on multi-state model | |
CN114290503B (en) | T beam steam health preserving generator power control method and device thereof | |
CN113901977A (en) | Deep learning-based power consumer electricity stealing identification method and system | |
CN104375478A (en) | Method and device for online predicting and optimizing product quality in steel rolling production process | |
CN110083096A (en) | A kind of silo intelligent monitoring method | |
CN107506862A (en) | A kind of online real-time estimate system and method for grinding particle size based on Internet of Things | |
CN111695730A (en) | ARIMA and RNN-based vertical mill vibration prediction method and device | |
CN114418177B (en) | New product material distribution prediction method based on digital twin workshops for generating countermeasure network | |
CN114757413A (en) | Bad data identification method based on time sequence series analysis coupling neural network prediction | |
CN101285816A (en) | Copper matte air refining procedure parameter soft sensing instrument and its soft sensing method | |
CN105259754B (en) | A kind of plate thickness intelligent control method based on Active Learning | |
CN106599391B (en) | Association vector machine soft measurement modeling method based on dynamic weighting of triangle angle values | |
CN107703752A (en) | A kind of Digit Control Machine Tool detection method based on fuzzy neural network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C05 | Deemed withdrawal (patent law before 1993) | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20130828 |