CN106709149A - Neural network-based method and system for predicting shapes of three-dimensional hearths of aluminum cells in real time - Google Patents
Neural network-based method and system for predicting shapes of three-dimensional hearths of aluminum cells in real time Download PDFInfo
- Publication number
- CN106709149A CN106709149A CN201611055395.0A CN201611055395A CN106709149A CN 106709149 A CN106709149 A CN 106709149A CN 201611055395 A CN201611055395 A CN 201611055395A CN 106709149 A CN106709149 A CN 106709149A
- Authority
- CN
- China
- Prior art keywords
- data
- burner hearth
- neural network
- aluminium
- hearth shape
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/23—Design optimisation, verification or simulation using finite element methods [FEM] or finite difference methods [FDM]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Control Of Steam Boilers And Waste-Gas Boilers (AREA)
- Manufacture Of Iron (AREA)
Abstract
The invention discloses a neural network-based method and system for predicting the shapes of three-dimensional hearths an aluminum cells in real time. The method comprises the following steps of: 1) obtaining aluminum cell modeling parameters; 2) constructing an aluminum cell heat transfer finite element model according to the aluminum cell modeling parameters, carrying out permutation and combination on different experimental parameters to form experimental groups, and respectively inputting each group of experimental parameters in the experimental groups into the aluminum cell heat transfer finite element model to carry out simulation calculation, so as to obtain a hearth shape and a cell shell temperature corresponding to each group experimental parameters; 3) constructing a BP neural network by taking the cell shell temperature, an electrolyte level and an aluminum level as input variables and taking the hearth shape as an output variable; 4) training the BP neural network on the basis of a simulation result in the step 2); and 5) measuring and obtaining the cell shell temperature, the electrolyte level and the aluminum level in real time, and inputting the cell shell temperature, the electrolyte level and the aluminum level into the BP neural network trained in the step 4), so as to obtain a finally predicted hearth shape. According to the method and system disclosed by the invention, the shapes of the three-dimensional hearths of the aluminum cells can be correctly and rapidly predicted.
Description
Technical field
The invention belongs to aluminium cell field, more particularly to a kind of aluminium cell three-dimensional burner hearth shape based on neutral net
Shape real-time predicting method and system.
Background technology
In aluminium cell, electric current produces Joule heat by the fused electrolyte for having a constant impedance, maintains the height of electrolytic cell
Temperature operation.And fused electrolyte cooled and solidified in contact side wall, forming one encloses stove side and stretches one's legs, and energy protective side wall does not receive high temperature
Melt attack.Suitable burner hearth shape is maintained, the horizontal current in aluminium liquid can be reduced, improve hydromagnetic stability in groove, had
Beneficial to the stable operation of production process.And irrational burner hearth shape can cause groove internal stability to deteriorate, current efficiency is reduced, sternly
Heavy even can occur bakie, so as to directly affect electrolysis safety in production and items technical-economic index.
Melt temperature now during aluminum electrolysis, is also difficult to stove close to 1000 DEG C and with highly corrosive in groove
Thorax shape carries out directly effective observation.In industrial production, special instrument is stretched into when being frequently in change poles, determined several
Location point speculates the burner hearth shape of remaining position.It is close the time required to measuring full tank furnace thorax shape because anode change cycle is more long
One month, when measuring, the burner hearth shape of last measurement position often had occurred and that change, and this method measuring point has
Limit, the stove side shape accuracy for measuring is poor, and site operation personnel can only be aided in judge situation in groove.
In the prior art, patent application " furnace edge shape of aluminium electrolytic tank on-line monitoring system " (application number：
201110439642.8) in, a kind of furnace edge shape of aluminium electrolytic tank on-line monitoring system is disclosed, on electrolytic cell (1) pot shell
Electrolysis tank furnace side (2) connects temperature probe (3), and temperature probe (3) is electrically connected with upper industrial computer (6).The patent application can be even
Continuous monitoring electrolyzer temperature, but both do not disclosed and carry out the specific method that furnace edge shape of aluminium electrolytic tank calculates analysis, more not right
Whether result of calculation is verified, it is impossible to determined whether groove internal furnace shape is regular, had the risk of bakie.
Therefore, it is necessary to design a kind of new aluminium cell three-dimensional burner hearth shape real-time predicting method and system.
The content of the invention
Technical problem solved by the invention is, in view of the shortcomings of the prior art, there is provided a kind of aluminium based on neutral net
Electrolytic cell three-dimensional burner hearth shape real-time predicting method and system, BP neural network is trained using simulation result data, from
And groove internal furnace shape can be rapidly and accurately predicted according to shell temperature.
Technical solution of the invention is as follows：
A kind of aluminium cell three-dimensional burner hearth shape real-time predicting method based on neutral net, comprises the following steps：
1) obtain aluminium cell modeling parameters, including each inner lining material attribute data and structural parameters, extraneous heat transfer boundary condition,
Technological parameter；
2) modeling parameters obtained according to step 1 build aluminium cell heat transfer FEM model；By different tests parameter
Different values carry out permutation and combination, constitute test group；Each group of test parameters in test group is input into aluminium cell respectively to pass
Hot FEM model, carries out simulation calculation, obtains the corresponding burner hearth shape of each group of test parameters and shell temperature data, is used for
Follow-up training；
3) BP neural network is built；
Choose the input variable of shell temperature, electrolyte level and aluminium level as the BP neural network；
Choose output variable of the burner hearth shape as the BP neural network；
4) BP neural network is trained, the BP neural network debugged is obtained；
Vi. from step 2) simulation result in obtain shell temperature data；Electrolyte level and aluminium water are obtained from slot control machine
Flat data；And all of shell temperature data, electrolyte level and aluminium horizontal data are divided into the first training set and the first survey
Examination collection；
Vii. from step 2) simulation result in obtain burner hearth shape；And all of burner hearth shape is divided into the second training
Collection and the second test set；
Viii. by the shell temperature data in first training set, electrolyte level and aluminium horizontal data and described
Burner hearth shape data in second training set carries out BP neural network training as the inputoutput data of the BP neural network,
Until BP neural network convergence【Mean square error is less than 10-3It is judged to convergence】；
Ix. the shell temperature data in first test set, electrolyte level and aluminium horizontal data are input into the receipts
BP neural network after holding back, exports the burner hearth shape data of prediction；
X. the burner hearth shape data of the prediction is compared with the burner hearth shape data in the second test set, inspection prediction effect
Really；If predictablity rate is higher than 90%, using current BP neural network as the BP neural network debugged, into next
Step；If predictablity rate is less than 90%, return to step I is repartitioned training set and test set and is trained；
5) in aluminum cell side and bottom pot shell arrangement temperature sensor, pot shell temperature data is measured in real time；While from
Slot control machine obtains electrolyte level and aluminium horizontal data；By the shell temperature data and electrolyte level and aluminium level of real-time measurement
Data input step 4) the middle BP neural network debugged, obtain the burner hearth shape data finally predicted.
Aluminium cell heat transfer FEM model and the method for solving set up in above-mentioned step 2, can select Xu Yujie, Lee
Jie, Yin Chenggang, et al.《Aluminium cell electric-thermal field intensity Coupling method computational methods》The model and method for solving of middle proposition.
Preferably, test parameters is one or more in following parameter in above-mentioned step 2：Electrolyte level, aluminium water
Flat, pole span height, current strength, mulch thickness, outside air temperature.
Preferably, above-mentioned step 3 builds BP neural network, including：
Determine the structure of BP neural network, including the hidden layer number of plies, each node layer number of input layer, hidden layer, output layer；
Determine neuron connection weight and neuron threshold value.
Preferably, the step 4) in also include：
Respectively to the shell temperature data in first training set and the first test set, electrolyte level and aluminium number of levels
According to being normalized；
The burner hearth shape data in second training set and the second test set is normalized respectively；
BP neural network training is carried out based on the data after normalization.
Preferably, the step 5) in, first to measurement pot shell temperature data in real time, from slot control machine obtain electrolyte level and
Aluminium horizontal data is normalized；Input step 4 again) in BP neural network debug, obtain the final burner hearth predicted
Shape data；Finally the burner hearth shape data of the final prediction to exporting carries out renormalization treatment, the burner hearth after being predicted
Shape data.
Preferably, in above-mentioned method, described temperature sensor is following one or more：Thermocouple, temperature-sensitive electricity
Resistance, resistance temperature detector, IC temperature sensors.
Shell temperature data to temperature sensor measurement carry out interpolation processing, to ask for missing data and anon-normal constant
According to substitution value.
A kind of aluminium cell three-dimensional burner hearth shape real-time estimate system based on neutral net, including：
Parameter acquiring unit, obtains each inner lining material attribute data of aluminium cell with structural parameters, extraneous heat transfer boundary condition, work
Skill parameter；
Simulation unit, the modeling parameters for being obtained according to step 1 build aluminium cell heat transfer FEM model；By difference
The different values of test parameters carry out permutation and combination, constitute test group；Each group of test parameters in test group is input into respectively
Aluminium cell conducts heat FEM model, carries out simulation calculation, obtains the corresponding burner hearth shape of each group of test parameters and pot shell temperature
Degrees of data, for follow-up training；
Neutral net construction unit, for building BP neural network；
Input/output variable chooses unit, for choosing shell temperature, electrolyte level and aluminium level as BP god
Through the input variable of network；Choose output variable of the burner hearth shape as the BP neural network；
First division unit, for from simulation result in obtain shell temperature data；Electrolyte water is obtained from slot control machine
Gentle aluminium horizontal data；And all of shell temperature data, electrolyte level and aluminium horizontal data are divided into the first training set
With the first test set；；
Second division unit, for obtaining burner hearth shape from simulation result；And all of burner hearth shape is divided into
Two training sets and the second test set；
Training unit, by the shell temperature data in first training set, electrolyte level and aluminium horizontal data and
Burner hearth shape data in second training set carries out BP neural network as the inputoutput data of the BP neural network
Training, until the BP neural network restrains；
Processing unit, by the input of the shell temperature data in first test set, electrolyte level and aluminium horizontal data
BP neural network after the convergence, exports the burner hearth shape data of prediction；
Verification unit, the burner hearth shape data of the prediction is compared with the burner hearth shape data in the second test set, inspection
Test prediction effect；If predictablity rate is higher than 90%, using current BP neural network as the BP neural network debugged；If
Predictablity rate is less than 90%, then control that the first division unit and the second division unit repartition training set and test set enters again
Row training；
Temperature detection and processing unit, including the temperature sensor of aluminum cell side and bottom pot shell is arranged in, it is used for
Pot shell temperature data is measured in real time；And interpolation processing is carried out to the shell temperature data to temperature sensor measurement, to ask for lacking
Lose the substitution value of data and improper data；
Predicting unit：The shell temperature data and electrolyte level of real-time measurement and the input of aluminium horizontal data have been debugged
BP neural network, obtains the burner hearth shape data finally predicted.
Preferably, test parameters is one or more in following parameter in the simulation unit of said system：Electrolyte water
Flat, aluminium level, pole span height, current strength, mulch thickness, outside air temperature.
Preferably, the neutral net construction unit of said system, including：
First determining module, the structure for determining BP neural network, including the hidden layer number of plies, it is input layer, hidden layer, defeated
Go out each node layer number of layer；
Second determining module, for determining neuron connection weight and neuron threshold value.
Preferably, said system also includes：
First normalized module, for being carried out to shell temperature data, electrolyte level and aluminium horizontal data respectively
Normalized；
Second normalized module, for being normalized to burner hearth shape data respectively；
Renormalization processing module, the burner hearth shape data for the final prediction to exporting carries out renormalization treatment,
Burner hearth shape data after being predicted.
Beneficial effect：
The inventive method possesses following advantage compared with existing burner hearth form measuring method：
(1) burner hearth shape is predicted based on the BP neural network of debugging completion, can be according to the result of test, with reference to electrolysis
Actual motion state, the burner hearth shape of each position in aluminium cell is rapidly and accurately obtained, so that it is guaranteed that aluminium cell
Safety in production.
(2) BP neural network is trained using the simulation result of different tests parameter permutation and combination, Neng Gouti
For enough amount of training data, it is ensured that the convergence of neutral net.
(3) missing data of temperature sensor and the substitution value of improper data are asked for using interpolation method, can be obtained
More accurate inputoutput data, effectively raises precision of prediction.
Brief description of the drawings
Fig. 1 is the schematic flow sheet of aluminium cell three-dimensional burner hearth shape real-time predicting method provided in an embodiment of the present invention；
Fig. 2 is the temperature of the part aluminium cell heat transfer FEM model determined according to the different values of different tests parameter
Distribution situation；Fig. 2 (a)~(d) is respectively according to 4 groups of profiling temperatures of the model of parameter determination, wherein A, B, C, D, E,
F, G, H, I represent different temperature values (DEG C) respectively, respectively equal to 117.362,217.084,316.806,416.528,
516.25、615.972、715.695、815.417、915.139；
Fig. 3 is electrolyzer temperature sensor installation position schematic diagram；
Fig. 4 is the schematic diagram of aluminium cell three-dimensional burner hearth shape real-time estimate system provided in an embodiment of the present invention.
Specific embodiment
The invention will be further described below in conjunction with the accompanying drawings, but protection scope of the present invention is not limited by embodiment.
Embodiment 1
A kind of aluminium cell three-dimensional burner hearth shape real-time predicting method based on neutral net, it is characterised in that including with
Lower step：
1) certain 420kA aluminium cell modeling parameters is obtained, including each inner lining material attribute data is changed with structural parameters, the external world
Heat condition, technological parameter, as shown in table 1；
Table 1 certain 420kA electrolytic cell key parameter
2) modeling parameters obtained according to step 1 build aluminium cell heat transfer FEM model.Meanwhile, in order to cover
The most of situation occurred in production, is easy to be subsequently used for training the burner hearth shape, it is necessary under the conditions of calculating different tests.Cause
This, permutation and combination is carried out by different tests parameter, constitutes test group；Each group of test parameters in test group is input into aluminium respectively
Electrolytic cell conducts heat FEM model, carries out simulation calculation, obtains the corresponding burner hearth shape of each group of test parameters and shell temperature
Data；
Aluminium cell heat transfer FEM model and the method for solving of above-mentioned foundation, from Xu Yujie, Lee's Jie, Yin Chenggang,
Et al.《Aluminium cell electric-thermal field intensity Coupling method computational methods》The model and method for solving of middle proposition.
The test parameters chosen in the present embodiment is current strength, mulch thickness, by current strength, mulch thickness
Different values input aluminium cells heat transfer FEM models, it is determined that aluminium cell conduct heat the part of FEM model test group
As shown in Figure 2：
The permutation and combination of the test parameters of table 1 (current strength, mulch thickness) is illustrated
3) BP neural network is built；
Including：
Determine the structure of BP neural network, including the hidden layer number of plies, each node layer number of input layer, hidden layer, output layer；
Determine neuron connection weight and neuron threshold value.
Choose the input variable of shell temperature, electrolyte level and aluminium level as the BP neural network；
Choose output variable of the burner hearth shape as the BP neural network；
I. the first training set and the first test set and by the shell temperature, electrolyte level and aluminium are divided horizontally into；Will
Burner hearth shape data is divided into the second training set and the second test set；
Ii. the input variable in first training set and the first test set is normalized respectively；It is right respectively
Burner hearth shape data in second training set and the second test set is normalized；
Before neural metwork training, in order to prevent neuron is exported saturation because net input absolute value is excessive, and then make
Weighed value adjusting into error surface flat region, it is necessary to be normalized to the data of training set and test set, to each instruction
Practice the data x of collection and test seti, i=1,2 ..., n, using following normalization formula：
Data after being normalized, input layer and output layer data are mapped between [0,1], this nondimensionalization side
Method, solves the problems, such as caused by dimension and numerical values recited difference.
Wherein, xiK gained is mapped to the data of original data area after () expression renormalization；xi' (k) represents normalization
Data afterwards；
Iii. by the burner hearth shape in the input variable in the first training set after normalized and second training set
Shape data carry out BP neural network training as the inputoutput data of the BP neural network, until mean square error function is less than
10-3, the BP neural network convergence；
Iv. the BP neural network input variable in the first test set after normalized being input into after the convergence, it is defeated
Go out the burner hearth shape data of prediction；
V. by the burner hearth shape data in the second test set after the burner hearth shape data of the prediction and normalized
Compare, check prediction effect；If prediction data accuracy rate is higher than 90%, into next step, less than 90% return to step 4,
Re-start division training；
4) in aluminum cell side and bottom pot shell arrangement temperature sensor, as shown in figure 3, measurement shell temperature number in real time
According to；Simultaneously electrolyte level and aluminium horizontal data are obtained from slot control machine；To the shell temperature data and electrolyte water of real-time measurement
Gentle aluminium horizontal data is input into the good BP neural network of above-mentioned debugging after being normalized, obtain the burner hearth shape finally predicted
Shape.The burner hearth shape data of the final prediction to exporting carries out renormalization treatment, the burner hearth shape data after being predicted.
Can utilize following renormalization treatment formula that model output is mapped into original data area：
xi(k)=x 'i(k)[max(xi)-min(xi)]+min(xi)
Wherein, xiK gained is mapped to the data of original data area after () expression renormalization；x′iK () represents normalization
Data afterwards,
In the present embodiment, processed by the normalization to data and renormalization, more accurate prediction data can be obtained,
Effectively raise precision of prediction.
The temperature sensor selected in the present embodiment is thermocouple.Due to originals such as signal acquisition exception, network communication failures
Cause, may cause some shortage of data in temperature sensor data, or some data variation exceptions, such as excessive or too small.It is right
In such abnormal data, it is necessary to be processed on the basis of qualitative analysis.In the present embodiment, temperature is asked for using interpolation method
The substitution value of degree sensor missing data and improper data.
The substitution value of temperature sensor missing data and improper data is asked for using interpolation method, it is possible to use below
Formula carry out：
x0(i)=x0(k)+[x0(j)-x0(k)]*(i-k)/(j-k)
Wherein：x0(k)、x0J () is known normal data, x0I () is the replacement of missing data or improper data；
k<i<j；
I represents the sequence number of missing data or abnormal data in data sequence, and j, k represent given data in data sequence
Sequence number, missing data or abnormal data are substituted using the inside interpolation of front and rear normal given data.
Shell temperature data such as following table that certain time point collects (after interpolation processing)：
Shell temperature data after the interpolation processing of table 2
The BP neural network model that data above input has been debugged, can obtain burner hearth shape data, as shown in the table：
The each position burner hearth thickness data that table 3 is calculated by BP neural network
The data of table 3 illustrate that the real-time predicting method and system provided by the present invention can rapidly and accurately obtain aluminium electricity
The real-time burner hearth thickness of each position of solution groove.
Claims (10)
1. a kind of aluminium cell three-dimensional burner hearth shape real-time predicting method based on neutral net, it is characterised in that including following
Step：
1) aluminium cell modeling parameters, including each inner lining material attribute data and structural parameters, extraneous heat transfer boundary condition, technique are obtained
Parameter；
2) modeling parameters obtained according to step 1 build aluminium cell heat transfer FEM model；By the difference of different tests parameter
Value carries out permutation and combination, constitutes test group；Each group of test parameters in test group is input into aluminium cell heat transfer respectively has
Limit meta-model, carries out simulation calculation, obtains the corresponding burner hearth shape of each group of test parameters and shell temperature data；
3) BP neural network is built；
Choose the input variable of shell temperature, electrolyte level and aluminium level as the BP neural network；
Choose output variable of the burner hearth shape as the BP neural network；
4) BP neural network is trained, the BP neural network debugged is obtained；
5) in aluminum cell side and bottom pot shell arrangement temperature sensor, pot shell temperature data is measured in real time；Simultaneously from groove control
Machine obtains electrolyte level and aluminium horizontal data；By the shell temperature data and electrolyte level and aluminium horizontal data of real-time measurement
Input step 4) the middle BP neural network debugged, obtain the burner hearth shape data finally predicted.
2. the aluminium cell three-dimensional burner hearth shape real-time predicting method based on neutral net according to claim 1, it is special
Levy and be, the step 4) specifically include following steps：
I. from step 2) simulation result in obtain shell temperature data；Electrolyte level and aluminium number of levels are obtained from slot control machine
According to；And all of shell temperature data, electrolyte level and aluminium horizontal data are divided into the first training set and the first test set；
Ii. from step 2) simulation result in obtain burner hearth shape；And by all of burner hearth shape be divided into the second training set and
Second test set；
Iii. by the shell temperature data in first training set, electrolyte level and aluminium horizontal data and second instruction
Practice the burner hearth shape data concentrated carries out BP neural network training as the inputoutput data of the BP neural network, until institute
State BP neural network convergence；
Iv. after the shell temperature data in first test set, electrolyte level and aluminium horizontal data being input into the convergence
BP neural network, export prediction burner hearth shape data；
V. the burner hearth shape data of the prediction is compared with the burner hearth shape data in the second test set, checks prediction effect；
If predictablity rate is higher than 90%, using current BP neural network as the BP neural network debugged, into next step；If
Predictablity rate is less than 90%, then return to step I, repartitions training set and test set and be trained.
3. the aluminium cell three-dimensional burner hearth shape real-time predicting method based on neutral net according to claim 1, it is special
Levy and be, the step 2) in test parameters be one or more in following parameter：Electrolyte level, aluminium level, pole span are high
Degree, current strength, mulch thickness and outside air temperature.
4. the aluminium cell three-dimensional burner hearth shape real-time predicting method based on neutral net according to claim 1, it is special
Levy and be, the step 3) BP neural network is built, including：
Determine the structure of BP neural network, including the hidden layer number of plies, each node layer number of input layer, hidden layer, output layer；
Determine neuron connection weight and neuron threshold value.
5. the aluminium cell three-dimensional burner hearth shape real-time predicting method based on neutral net according to claim 1, it is special
Levy and be, the step 4) in also include：
The shell temperature data in first training set and the first test set, electrolyte level and aluminium horizontal data are entered respectively
Row normalized；
The burner hearth shape data in second training set and the second test set is normalized respectively；
BP neural network training is carried out based on the data after normalization.
6. the aluminium cell three-dimensional burner hearth shape real-time predicting method based on neutral net according to claim 1, it is special
Levy and be, the step 5) in,
From first to measurement pot shell temperature data in real time, from slot control machine acquisition electrolyte level and aluminium horizontal data being normalized
Reason；Input step 4 again) in BP neural network debug, obtain the final burner hearth shape data predicted；Finally output to most
The burner hearth shape data of prediction carries out renormalization treatment eventually, the burner hearth shape data after being predicted.
7. the three-dimensional burner hearth shape of the aluminium cell based on neutral net according to any one of claim 1~6 is pre- in real time
Survey method, it is characterised in that the step 5) also include carrying out interpolation processing to the shell temperature data of temperature sensor measurement,
To ask for the substitution value of missing data and improper data.
8. a kind of aluminium cell three-dimensional burner hearth shape real-time estimate system based on neutral net, it is characterised in that including：
Parameter acquiring unit, obtains each inner lining material attribute data of aluminium cell with structural parameters, extraneous heat transfer boundary condition, technique ginseng
Number；
Simulation unit, the modeling parameters for being obtained according to step 1 build aluminium cell heat transfer FEM model；By different tests
The different values of parameter carry out permutation and combination, constitute test group；Each group of test parameters in test group is input into aluminium electricity respectively
Solution groove heat transfer FEM model, carries out simulation calculation, obtains the corresponding burner hearth shape of each group of test parameters and shell temperature number
According to；
Neutral net construction unit, for building BP neural network；
Input/output variable chooses unit, for choosing shell temperature, electrolyte level and aluminium level as the BP nerve nets
The input variable of network；Choose output variable of the burner hearth shape as the BP neural network；
First division unit, for from simulation result in obtain shell temperature data；From slot control machine obtain electrolyte level and
Aluminium horizontal data；And all of shell temperature data, electrolyte level and aluminium horizontal data are divided into the first training set and
One test set；；
Second division unit, for obtaining burner hearth shape from simulation result；And all of burner hearth shape is divided into the second instruction
Practice collection and the second test set；
Training unit, by the shell temperature data in first training set, electrolyte level and aluminium horizontal data and described
Burner hearth shape data in second training set carries out BP neural network training as the inputoutput data of the BP neural network,
Until BP neural network convergence；
Processing unit, the input of the shell temperature data in first test set, electrolyte level and aluminium horizontal data is described
BP neural network after convergence, exports the burner hearth shape data of prediction；
Verification unit, the burner hearth shape data of the prediction is compared with the burner hearth shape data in the second test set, and inspection is pre-
Survey effect；If predictablity rate is higher than 90%, using current BP neural network as the BP neural network debugged；If prediction
Accuracy rate is less than 90%, then control that the first division unit and the second division unit repartition training set and test set is instructed again
Practice；
Temperature detection and processing unit, including the temperature sensor of aluminum cell side and bottom pot shell is arranged in, for real-time
Measurement pot shell temperature data；And interpolation processing is carried out to the shell temperature data to temperature sensor measurement, to ask for missing number
According to this and improper data substitution value；
Predicting unit：The BP god that the shell temperature data and electrolyte level of real-time measurement and the input of aluminium horizontal data have been debugged
Through network, the burner hearth shape data finally predicted is obtained.
9. the aluminium cell three-dimensional burner hearth shape real-time estimate system based on neutral net according to claim 8, it is special
Levy and be, the neutral net construction unit, including：
First determining module, the structure for determining BP neural network, including the hidden layer number of plies, input layer, hidden layer, output layer
Each node layer number；
Second determining module, for determining neuron connection weight and neuron threshold value.
10. the aluminium cell three-dimensional burner hearth shape real-time estimate system based on neutral net according to claim 8, it is special
Levy and be,
First normalized module, for carrying out normalizing to shell temperature data, electrolyte level and aluminium horizontal data respectively
Change is processed；
Second normalized module, for being normalized to burner hearth shape data respectively；
Renormalization processing module, the burner hearth shape data for the final prediction to exporting carries out renormalization treatment, obtains
Burner hearth shape data after prediction.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611055395.0A CN106709149B (en) | 2016-11-25 | 2016-11-25 | A kind of aluminium cell three-dimensional burner hearth shape real-time predicting method neural network based and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611055395.0A CN106709149B (en) | 2016-11-25 | 2016-11-25 | A kind of aluminium cell three-dimensional burner hearth shape real-time predicting method neural network based and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106709149A true CN106709149A (en) | 2017-05-24 |
| CN106709149B CN106709149B (en) | 2019-11-19 |
Family
ID=58933935
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611055395.0A Active CN106709149B (en) | 2016-11-25 | 2016-11-25 | A kind of aluminium cell three-dimensional burner hearth shape real-time predicting method neural network based and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106709149B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109283207A (en) * | 2018-11-02 | 2019-01-29 | 中南大学 | The detection device and method of a kind of simulation electrolytic bath of aluminium side growth course |
| CN109732450A (en) * | 2019-02-27 | 2019-05-10 | 重庆理工大学 | A kind of blade polishing processing method neural network based |
| CN116288532A (en) * | 2023-03-13 | 2023-06-23 | 赛富能科技(深圳)有限公司 | Method and equipment for monitoring electrolytic tank |
| CN117253568A (en) * | 2023-11-15 | 2023-12-19 | 沈阳恒泰鑫源精铸耐材有限公司 | Coating process optimization method and system for preparing yttrium oxide crucible |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2010011918A2 (en) * | 2008-07-24 | 2010-01-28 | University Of Cincinnati | Methods for prognosing mechanical systems |
| CN102517610A (en) * | 2011-12-26 | 2012-06-27 | 贵阳铝镁设计研究院有限公司 | On-line monitoring system of furnace edge shape of aluminium electrolytic tank |
| US20130304680A1 (en) * | 2012-05-10 | 2013-11-14 | Bp Exploration Operating Company Limited | Predictive corrosion coupons from data mining |
| CN106021760A (en) * | 2016-05-27 | 2016-10-12 | 中南大学 | Aluminum electrolysis cell magnetic field finite element calculation method based on ANSYS |
-
2016
- 2016-11-25 CN CN201611055395.0A patent/CN106709149B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2010011918A2 (en) * | 2008-07-24 | 2010-01-28 | University Of Cincinnati | Methods for prognosing mechanical systems |
| CN102517610A (en) * | 2011-12-26 | 2012-06-27 | 贵阳铝镁设计研究院有限公司 | On-line monitoring system of furnace edge shape of aluminium electrolytic tank |
| US20130304680A1 (en) * | 2012-05-10 | 2013-11-14 | Bp Exploration Operating Company Limited | Predictive corrosion coupons from data mining |
| CN106021760A (en) * | 2016-05-27 | 2016-10-12 | 中南大学 | Aluminum electrolysis cell magnetic field finite element calculation method based on ANSYS |
Non-Patent Citations (2)
| Title |
|---|
| 徐宇杰 等: "铝电解槽电-热场强耦合建模计算方法", 《中国有色金属学报》 * |
| 苏少航: "基于人工神经网络的焊接熔池形状预测", 《江苏工业学院学报》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109283207A (en) * | 2018-11-02 | 2019-01-29 | 中南大学 | The detection device and method of a kind of simulation electrolytic bath of aluminium side growth course |
| CN109283207B (en) * | 2018-11-02 | 2020-09-29 | 中南大学 | Detection device and method for simulating growth process of furnace side of aluminum electrolytic cell |
| CN109732450A (en) * | 2019-02-27 | 2019-05-10 | 重庆理工大学 | A kind of blade polishing processing method neural network based |
| CN116288532A (en) * | 2023-03-13 | 2023-06-23 | 赛富能科技(深圳)有限公司 | Method and equipment for monitoring electrolytic tank |
| CN117253568A (en) * | 2023-11-15 | 2023-12-19 | 沈阳恒泰鑫源精铸耐材有限公司 | Coating process optimization method and system for preparing yttrium oxide crucible |
| CN117253568B (en) * | 2023-11-15 | 2024-01-26 | 沈阳恒泰鑫源精铸耐材有限公司 | Coating process optimization method and system for preparing yttrium oxide crucible |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106709149B (en) | 2019-11-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106709149B (en) | A kind of aluminium cell three-dimensional burner hearth shape real-time predicting method neural network based and system | |
| CN106352542B (en) | A kind of prediction technique of storage-type electric water heater residue bathing time | |
| CN112200433A (en) | Power plant thermal performance analysis and optimization system | |
| CN105137358B (en) | The SOC/SOH Forecasting Methodologies of power battery based on big data self-study mechanism | |
| CN110045594B (en) | Intelligent management and control system and method for predicting state risk of four tubes of boiler | |
| CN104767482B (en) | A kind of photovoltaic module is aging and short trouble inline diagnosis method | |
| CN106777984A (en) | A kind of method that photovoltaic array Working state analysis and fault diagnosis are realized based on density clustering algorithm | |
| CN107886171B (en) | PMU data-based breaker state online diagnosis method and system | |
| CN103984287A (en) | Numerically-controlled machine tool thermal error compensation grey neural network modeling method | |
| CN106126944B (en) | A kind of power transformer top-oil temperature interval prediction method and system | |
| CN106059496A (en) | Method and system for monitoring performance and identifying faults of array of photovoltaic assembly | |
| CN105320123A (en) | Measurement and control equipment health state evaluating system and method based on online monitored data | |
| CN116595327B (en) | Sluice deformation monitoring data preprocessing system and method | |
| CN101319256A (en) | Intelligent monitoring method for cooling wall of blast furnace | |
| CN107393616A (en) | A kind of nuclear power plant's reactor transparent monitoring system and method | |
| CN104914854B (en) | Industrial process fault diagnosis method based on KPCA | |
| CN105400951B (en) | A kind of humidification hybrid control system and its control method for annealing process of silicon steel | |
| CN113533622A (en) | Coal quality prediction method for coal mill based on neural network | |
| CN108536968A (en) | A kind of regulation and control method of stress and deformation in welding process | |
| CN106289754B (en) | A kind of fired power generating unit component capabilities variation determination method based on Statistical Control Technique | |
| CN104831317B (en) | Method for judging abnormal anode current of aluminum electrolysis cell | |
| CN104697585A (en) | Cooling tower tower-barrel strength predicating control system and method thereof | |
| CN109655483A (en) | A kind of material microstructure defect inspection method based on deep learning algorithm | |
| CN103647677B (en) | Power communication network state detection method | |
| CN111044176A (en) | Method for monitoring temperature abnormity of generator |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |