CN108984813A - Aluminium electroloysis modeling and optimization method based on recurrent neural network Yu angle preference - Google Patents
Aluminium electroloysis modeling and optimization method based on recurrent neural network Yu angle preference Download PDFInfo
- Publication number
- CN108984813A CN108984813A CN201810193126.3A CN201810193126A CN108984813A CN 108984813 A CN108984813 A CN 108984813A CN 201810193126 A CN201810193126 A CN 201810193126A CN 108984813 A CN108984813 A CN 108984813A
- Authority
- CN
- China
- Prior art keywords
- preference
- aluminium
- particle
- function
- neural network
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- 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
Abstract
The aluminium electroloysis modeling and optimization method based on recurrent neural network Yu angle preference that the present invention provides a kind of.First, aluminum electrolysis process is modeled using recurrent neural network, then policymaker sets expectation target value, introduce A-dominance preference administration method, production process model is optimized in conjunction with multi-target quantum particle swarm algorithm, most met the desired optimizing decision variable of policymaker and corresponding current efficiency, tank voltage, perfluoro-compound discharge amount and ton aluminium energy consumption.MQPSO algorithm do not need to be intersected, mutation operation, only simplest location updating step, therefore cataloged procedure is simple, and has strong ability of searching optimum, and the integrality of the optimal value of preference, meets policymaker's demand during Evolution of Population easy to accomplish.The optimal value of technological parameter during aluminum electrolysis is determined using this method, can effectively improve current efficiency, reduces tank voltage, is reduced greenhouse gas emissions, is achieved energy-saving and emission reduction purposes.
Description
Technical field
The invention belongs to optimum control fields, and in particular to a kind of aluminium electroloysis based on recurrent neural network Yu angle preference
Modeling and optimization method.
Background technique
Environment-friendly type aluminum electrolysis process is all very challenging for a long time, in Aluminium Industry, final goal
It is on the basis of electrolytic cell even running, improves current efficiency, reduces tank voltage and reduce perfluoro-compound, reduce ton aluminium energy
The discharge amount of consumption.However, aluminium cell parameter is more, and show non-linear, strong coupling between parameter, gives aluminum electrolysis
Process model building brings larger difficulty, and recurrent neural network has very strong non-linear mapping capability, is suitable for solving non-thread
Property system modelling problem, new thinking is provided for aluminum electrolysis process model building.And for four targets, while realizing then non-
It is often difficult, because target has the phenomenon that conflict between each other, the preference information of policymaker, setting expectation mesh can be introduced
Mark, is adjusted flexibly the weight between different target, carries out variable optimization using preference A-PMQPSO optimization algorithm.A-PMQPSO is
On the basis of MQPSO, A administration method is introduced.MQPSO is a kind of multi-objective optimization algorithm of classics, and the algorithm is simple, operation is fast
Degree is fast, evolutionary process can be described directly with equation, thus is widely used in multiple fields.
Summary of the invention
It is an object of the invention to overcome the deficiencies of the prior art and provide one kind based on recurrent neural network and angle preference
Aluminium electroloysis modeling and optimization method, to solve in the prior art during aluminum electrolysis because optimal procedure parameters can not be obtained
Caused by huge energy consumption, low efficiency and the technical issues of serious pollution environment, and at the same time decisionmaker's preference letter can be introduced
Breath realizes that dynamic flexible adjusts the purpose of preference weight between each target.
The object of the present invention is achieved like this:
A kind of aluminium electroloysis modeling and optimization method based on recurrent neural network Yu angle preference, includes the following steps:
S1: selection constitutes decision to current efficiency, tank voltage and the influential control parameter of perfluoro-compound discharge amount and becomes
Amount, decision variable X=[x1,x2,···,xM], M is the number of selected control parameter;
S2: selected aluminium electrolytic industry scene acquires N group decision variable X1,X2,···,XNAnd its corresponding current efficiency
y1,y2,···,yN, tank voltage z1,z2,···,zNAnd perfluoro-compound discharge amount s1,s2,···,sNWith ton aluminium energy
Consume c1,c2,···,cNFor data sample, with each group of decision variable XiAs input, with corresponding current efficiency yi, slot electricity
Press ziAnd perfluoro-compound discharge amount siWith ton aluminium energy consumption ciAs output, sample is trained using recurrent neural network, is examined
It tests, establishes four aluminium cell production process models;
S3: using the preference multi-target quantum particle swarm algorithm dominated based on A, according to the preset desired value of policymaker
It is as a reference point, the stringent partial ordering relation dominated based on A is established, four production process models resulting to step S2 carry out excellent
Change, obtains one group and most meet the desired decision variable X of policymakerbestAnd its corresponding current efficiency ybest, tank voltage zbestAnd
Perfluoro-compound discharge amount sbestWith ton aluminium energy consumption cbest;
S4: according to the resulting optimizing decision variable X of step S3bestIn control parameter come it is selected in rate-determining steps S2
Aluminium electrolytic industry scene, reaches the purpose of energy-saving and emission-reduction consumption reduction.
Preferably, in step S1, the control parameter includes potline current, blanking number, molecular proportion, aluminum yield, aluminum water
Flat, electrolyte level, bath temperature.
Preferably, in step S2, using current efficiency as output, aluminium cell production process model, input layer are established
Using 10 neuron nodes, hidden layer uses 15 neuron nodes, and output layer uses 1 neuron node, and input layer arrives
Transmission function is Tansig function between hidden layer, and hidden layer to the function between output layer is Purelin function, sample training
When the number of iterations be 1000.
Preferably, in step S2, using tank voltage as output, aluminium cell production process model is established, input layer is adopted
With 10 neuron nodes, hidden layer uses 15 neuron nodes, and output layer uses 1 neuron node, and input layer is to hidden
Hiding transmission function between layer is Logsig function, and hidden layer to the function between output layer is Purelin function, when sample training
The number of iterations be 1000.
Preferably, in step S2, using perfluoro-compound discharge amount as output, aluminium cell production process model is established,
Input layer uses 10 neuron nodes, and hidden layer uses 15 neuron nodes, and output layer uses 1 neuron node, defeated
Entering layer to transmission function between hidden layer is Logsig function, and hidden layer to the function between output layer is Purelin function, sample
The number of iterations when this training is 1000.
Preferably, in step S2, using ton aluminium energy consumption as output, aluminium cell production process model, input layer are established
Using 10 neuron nodes, hidden layer uses 15 neuron nodes, and output layer uses 1 neuron node, and input layer arrives
Transmission function is Tansig function between hidden layer, and hidden layer to the function between output layer is Purelin function, sample training
When the number of iterations be 1000.
Preferably, step S3 the following steps are included:
S31: the preference relation dominated according to A evaluates the fitness of each particle, and according to superiority and inferiority to individual optimal value and
Global optimum is replaced;
S311: initialization system parameter, including population scale R, maximum number of iterations T generate n particle x at random1,
x2,···,xn, enable external archival collection Q for sky;
S312: policymaker sets preference angle [alpha] and preference intended reference point r (yp,zp,sp,cp), the preference target ginseng
Examination point includes the desired value of current efficiency, tank voltage, perfluoro-compound discharge amount and ton aluminium four targets of energy consumption;
S313: for each individual x, its fitness and its angle with reference point reference line are calculated:
Wherein, fjIt (x) is fitness value of the individual x in jth target;
S314: being based on angle information on object space, preference zone is divided, if θ (r, x) < α, the individual are in
Preference zone;Otherwise it is in non-preference zone;
S315: judge any two individual xiWith xkBetween superiority and inferiority relationship, including following situations:
Work as xiWith xkWhen being in preference zone or non-preference zone simultaneously, if xiPareto dominates xk, then it is assumed that xiIt is more excellent,
If mutually Pareto is not dominated, then it is assumed that the two is suitable;
Work as xiIn preference zone, xkWhen in non-preference zone, if xiPareto dominates xkOr xiWith xkMutually not
Pareto is dominated, then it is assumed that xiBetter than xk, i.e. xiA dominates xk;
S316: the individual history optimal location pbest of each particle is determinedi, in system initialization, individual history is optimal
Position is set as the initial position x of the particlei;After next iteration, based on the A dominance relation that S315 is proposed, to the new of particle
Position xiWith pbestiSuperiority and inferiority comparison is carried out, outstanding person saves as pbesti;
S317: updating external archival collection Q, is added to the particle that non-A is dominated between each other in population and achieves collection Q, deletes quilt
The particle of domination;
S318: randomly choosed in external archival collection Q using press mechanism and Tabu search algorithm a particle as it is global most
Excellent position;
S32: Population Regeneration:
S321: the position of more new particle itself, wherein particle position more new formula are as follows:
Wherein: i (i=1,2 ..., n) represents i-th of particle, and n is population scale;J (j=1,2 ..., M) represent particle
Jth dimension, M are search space dimension;T is evolutionary generation;And uijIt (t) is equally distributed random in [0,1] section
Number;xij(t),pbestij(t) and γij(t) it is optimal that the current location of particle i, individual history when evolutionary generation is t are respectively indicated
Position and attractor position;gbestj(t) it respectively indicates global optimum position when evolutionary generation is t with mbest (t) and is averaged most
Good position;α indicates expansion-contraction factor;
S322: judging whether current globally optimal solution meets condition or whether the number of iterations reaches maximum number of iterations T,
If it is, exporting current globally optimal solution, otherwise, the S321 that gos to step is computed repeatedly, until current global optimum
Solution meets condition or the number of iterations reaches maximum number of iterations T.
By adopting the above-described technical solution, the present invention builds aluminum electrolysis process using recurrent neural network
Mould, then policymaker sets expectation target value, A-dominance preference administration method is introduced, in conjunction with multi-target quantum population
Algorithm optimizes production process model, is most met the desired optimizing decision variable of policymaker and corresponding electric current
Efficiency, tank voltage, perfluoro-compound discharge amount and ton aluminium energy consumption.MQPSO algorithm do not need to be intersected, mutation operation, only most
Simple location updating step, therefore cataloged procedure is simple, and has strong ability of searching optimum, Evolution of Population mistake easy to accomplish
The integrality of the optimal value of preference, meets policymaker's demand in journey.Technique is joined during determining aluminum electrolysis using this method
Several optimal values can effectively improve current efficiency, reduce tank voltage, reduce greenhouse gas emissions, reach the mesh of energy-saving and emission-reduction
's.
Detailed description of the invention
Fig. 1 is flow chart of the method for the present invention;
Fig. 2 is CF4 forecasting of discharged quantity result figure;
Fig. 3 is CF4 forecasting of discharged quantity Error Graph
Fig. 4 is current efficiency prediction result figure;
Fig. 5 is current efficiency prediction-error image;
Fig. 6 is tank voltage forecasting of discharged quantity result figure;
Fig. 7 is tank voltage forecasting of discharged quantity Error Graph;
Fig. 8 is ton aluminium energy consumption prediction result figure;
Fig. 9 is ton aluminium energy consumption prediction-error image.
Specific embodiment
As shown in Figure 1, a kind of aluminium electroloysis modeling and optimization method based on recurrent neural network Yu angle preference, including such as
Lower step:
S1: selection constitutes decision to current efficiency, tank voltage and the influential control parameter of perfluoro-compound discharge amount and becomes
Amount, decision variable X=[x1,x2,···,xM], M is the number of selected control parameter.
The present embodiment is by statistics aluminum electrolysis in the process to current efficiency, tank voltage and perfluoro-compound discharge amount
With the influential original variable of ton aluminium energy consumption, and therefrom determine to current efficiency, tank voltage and perfluoro-compound discharge amount and ton
The big parameter of aluminium energy consumption is as decision variable X;By being counted to measurement parameter during actual industrial production, obtain
To current efficiency, tank voltage and perfluoro-compound discharge amount and the maximum variable of ton aluminium energy consumption are as follows: potline current x1, blanking number
x2, molecular proportion x3, aluminum yield x4, the flat x of aluminum water5, electrolyte level x6, bath temperature x7Totally 7 variables;
S2: selected aluminium electrolytic industry scene acquires N group decision variable X1,X2,···,XNAnd its corresponding current efficiency
y1,y2,···,yN, tank voltage z1,z2,···,zNAnd perfluoro-compound discharge amount s1,s2,···,sNWith ton aluminium energy
Consume c1,c2,···,cNFor data sample, with each group of decision variable XiAs input, with corresponding current efficiency yi, slot electricity
Press ziAnd perfluoro-compound discharge amount siWith ton aluminium energy consumption ciAs output, sample is trained using recurrent neural network, is examined
It tests, establishes four aluminium cell production process models, the recurrent neural network includes input layer, hidden layer, output layer.
Four aluminium cells production process model includes:
For the production process model constructed by the current efficiency, input layer uses 10 neuron nodes, hides
Layer uses 15 neuron nodes, and output layer uses 1 neuron node, and input layer is to transmission function between hidden layer
Tansig function, hidden layer to the function between output layer are Purelin function, and the number of iterations when sample training is 1000.
For the production process model constructed by the tank voltage, input layer uses 10 neuron nodes, hidden layer
Using 15 neuron nodes, output layer uses 1 neuron node, and input layer to transmission function between hidden layer is Logsig
Function, hidden layer to the function between output layer are Purelin function, and the number of iterations when sample training is 1000.
For the production process model constructed by the perfluoro-compound discharge amount, aluminium cell production process model is established,
Its input layer uses 10 neuron nodes, and hidden layer uses 15 neuron nodes, and output layer uses 1 neuron node,
Input layer is Logsig function to transmission function between hidden layer, and hidden layer to the function between output layer is Purelin function,
The number of iterations when sample training is 1000.
For the production process model constructed by the ton aluminium energy consumption, aluminium cell production process model is established, is inputted
Layer uses 10 neuron nodes, and hidden layer uses 15 neuron nodes, and output layer uses 1 neuron node, input layer
It is Tansig function to transmission function between hidden layer, hidden layer to the function between output layer is Purelin function, sample instruction
The number of iterations when practicing is 1000.
In the present embodiment in order to meet modeling requirement, the recurrent neural network further includes associated layers.
In the present embodiment, the 223# slot electrolytic cell in Chongqing Tiantai Aluminium Industry Co., Ltd. 170KA series electrolytic cell is acquired
Whole year production data in 2013 and 40 day datas before 2014 amount to 405 groups of data, wherein whole year production data in 2013
As modeling training sample, 40 groups of data in 2014 are as test sample.Data sample is as shown in table 1 below.
1 data sample of table
Sample number | 1 | 2 | 3 | 4 | …… |
x1 | 1683 | 1682 | 1686 | 1746 | …… |
x2 | 624 | 716 | 625 | 743 | …… |
x3 | 2.52 | 2.52 | 2.51 | 2.46 | …… |
x4 | 1234 | 1230 | 1234 | 1235 | …… |
x5 | 18.5 | 16.5 | 17.5 | 20 | …… |
x6 | 14 | 14 | 15 | 16 | …… |
x7 | 942 | 938 | 946 | 942 | …… |
y1 | 94.65 | 94.66 | 94.43 | 93.22 | …… |
y2 | 3721 | 3720 | 3725 | 3717 | …… |
y3 | 4.25 | 4.84 | 4.01 | 4.15 | …… |
y4 | 12354.3 | 12316.4 | 12283.1 | 12747.2 | …… |
In recurrent neural network design, since there are recursive signal, network state is changed with time, therefore
In addition to the number of hidden nodes, learning rate similarly affects the stability and accuracy of neural network model, is that neural network is set
Heavy difficult point in meter.
The setting of the number of nodes of hidden layer is obtained by trial and error procedure:
In formula, p is hidden neuron number of nodes, and n is input layer number, and m is output layer neuron number, k 1-10
Between constant.
Best learning rate value are as follows:
The setting parameter of recurrent neural network is as shown in table 2 below in this example.
Parameter is arranged in 2 recurrent neural network of table
Objective function | Current efficiency | Tank voltage | Perfluoro-compound discharge amount | Ton aluminium energy consumption |
The number of iterations | 1000 | 1000 | 1000 | 1000 |
Hidden layer transmission function | Tansig | Logsig | Logsig | Tansig |
Output layer transmission function | Purelin | Purelin | Purelin | Purelin |
Node in hidden layer | 13 | 12 | 12 | 13 |
It is carried out in the training process of neural network essentially according to following steps:
X is setk=[xk1,xk2,···,xkM] (k=1,2, N) and it is input vector, N is training sample
Number,
Weight when for the g times iteration between input layer M and hidden layer I
Vector, WJP(g) be the g times iteration when hidden layer J and output layer P between weighted vector be Yk(g)=[yk1(g),
yk2(g),···,ykP(g)] (k=1,2, N) be the g times iteration when network reality output, dk=[dk1,
dk2,···,dkP] (k=1,2, N) it is desired output;
Aluminum electrolysis process model is established in step S2 to specifically comprise the following steps:
S21: initialization, the number of iterations g initial value are set as 0, WMI(0)、WJPIt (0) is for the random value in (0,1) section;
S22: stochastic inputs sample Xk
S23: to input sample Xk, the input signal and output signal of every layer of neuron of forward calculation recurrent neural network
S24: according to desired output dkWith reality output Yk(g), error E (g) is calculated;
Whether S25: error in judgement E (g) meet the requirements, and is such as unsatisfactory for, then enters step S26, such as meets, then enters step
S29;
S26: judging whether the number of iterations g+1 is greater than maximum number of iterations, such as larger than, then enters step S29, otherwise, into
Enter step S27;
S27: to input sample XkThe partial gradient of every layer of neuron of retrospectively calculate;
Network exports node layer error are as follows: e (k)=d (k)-y (k), e (k) are network desired output, and y (k) is that network is real
Border output.
By calculating output node layer error to the weight change rate of each layer are as follows:
Wherein βij(0)=0;I=1,2, n1;J=1,2, n0。
δi(0)=0;I=1,2, n1。
WhereinRespectively indicate the input and output of i-th of node of hidden layer;n0、n1Respectively output layer
And node in hidden layer;Respectively indicate associated layers, output layer, hidden layer weight.
S28: network weight modified computing formulae are as follows:
Wherein w (k) can be
W (k) can represent the weight of output layer, hidden layer or input layer in formula, and η is learning rate, enable g=g+1, jump to
Step S23;
S29: judging whether to complete all training samples, if it is, completing modeling, otherwise, continues to go to step
S22。
By above-mentioned cyclic process, recurrent neural networks prediction effect can be obtained as shown in Fig. 2,3,4,5,6,7,8,9.It is excellent
The foundation for changing model is the basis of aluminum electrolysis process optimization, and model accuracy directly affects optimum results.By to Fig. 2,3,
4,5,6,7,8,9 analyses are it is found that through recurrent neural network training, and the largest prediction error of current efficiency is 0.41%, tank voltage
Largest prediction error be 0.08%, carbon tetrafluoride CF4 forecasting of discharged quantity error -1.20%, ton aluminium energy consumption prediction error be
0.81%, model prediction accuracy is high, meets modeling demand.
S3: introducing A-dominance preference administration method, is calculated using the preference multi-target quantum population dominated based on A
Method, in conjunction with multi-target quantum particle swarm algorithm (MQPSO), i.e. A-PMQPSO algorithm, according to the preset desired value of policymaker
(reference point) establishes the stringent partial ordering relation dominated based on A, and four production process models resulting to step S2 optimize,
It obtains one group and most meets the desired decision variable X of policymakerbestAnd its corresponding current efficiency ybest, tank voltage zbestAnd it is complete
Fluoride emission amount sbestWith ton aluminium energy consumption cbest。
On the basis of aluminum electrolysis process model, it is carried out within the scope of each decision variable using A-PMQPSO algorithm
Optimization, each specific variation range of variable are as shown in table 3.
Each variable-value range of table 3
In step S3, the A-PMQPSO algorithm the following steps are included:
S31: the preference relation dominated according to A evaluates the fitness of each particle, and according to superiority and inferiority to individual optimal value and
Global optimum is replaced;
S311: initialization system parameter, including population scale R, maximum number of iterations T generate n particle x at random1,
x2,···,xn, enable external archival collection Q for sky;
S312: policymaker sets preference angle [alpha] and preference intended reference point r (yp,zp,sp,cp), the preference target ginseng
Examination point includes the desired value of current efficiency, tank voltage, perfluoro-compound discharge amount and ton aluminium four targets of energy consumption;
S313: for each individual x, its fitness and its angle with reference point reference line are calculated:
Wherein, fjIt (x) is fitness value of the individual x in jth target;
S314: being based on angle information on object space, preference zone is divided, if θ (r, x) < α, the individual are in
Preference zone;Otherwise it is in non-preference zone;
S315: judge any two individual xiWith xkBetween superiority and inferiority relationship, including following situations:
Work as xiWith xkWhen being in preference zone or non-preference zone simultaneously, if xiPareto dominates xk, then it is assumed that xiIt is more excellent,
If mutually Pareto is not dominated, then it is assumed that the two is suitable;
Work as xiIn preference zone, xkWhen in non-preference zone, if xiPareto dominates xkOr xiWith xkMutually not
Pareto is dominated, then it is assumed that xiBetter than xk, i.e. xiA dominates xk;
S316: the individual history optimal location pbest of each particle is determinedi, in system initialization, individual history is optimal
Position is set as the initial position x of the particlei;After next iteration, based on the A dominance relation that S315 is proposed, to the new of particle
Position xiWith pbestiSuperiority and inferiority comparison is carried out, outstanding person saves as pbesti;
S317: updating external archival collection Q, is added to the particle that non-A is dominated between each other in population and achieves collection Q, deletes quilt
The particle of domination;
S318: randomly choosed in external archival collection Q using press mechanism and Tabu search algorithm a particle as it is global most
Excellent position;
S32: Population Regeneration:
S321: the position of more new particle itself, wherein particle position more new formula are as follows:
Wherein: i (i=1,2 ..., n) represents i-th of particle, and n is population scale;J (j=1,2 ..., M) represent particle
Jth dimension, M are search space dimension;T is evolutionary generation;And uijIt (t) is equally distributed random in [0,1] section
Number;xij(t),pbestij(t) and γij(t) it is optimal that the current location of particle i, individual history when evolutionary generation is t are respectively indicated
Position and attractor position;gbestj(t) it respectively indicates global optimum position when evolutionary generation is t with mbest (t) and is averaged most
Good position;α indicates expansion-contraction factor;
S322: judging whether current globally optimal solution meets condition or whether the number of iterations reaches maximum number of iterations T,
If it is, exporting current globally optimal solution, otherwise, the S321 that gos to step is computed repeatedly, until current global optimum
Solution meets condition or the number of iterations reaches maximum number of iterations T.
Aluminum electrolysis process is optimized through the above steps can obtain 100 groups of optimal decision variables with it is corresponding defeated
It is worth out, chooses and be wherein listed in the table below in 4 for most reasonable 3 groups.
4 optimized producing parameter of table
y1 | y2 | y3 | y4 | x1 | x2 | x3 | x4 | x5 | x6 | x7 |
99.14 | 3635 | 3.65 | 10835.15 | 1649 | 628 | 2.54 | 1210 | 16.5 | 14.5 | 942 |
98.13 | 3682 | 3.59 | 11527.21 | 1652 | 626 | 2.38 | 1200 | 17.5 | 15 | 925 |
95.37 | 3602 | 3.68 | 10478.32 | 1674 | 617 | 2.47 | 1095 | 17.5 | 15.5 | 935 |
It is compared using the average value of optimal operating parameter and annual record in 2013 it is found that current efficiency improves
3.89%, tank voltage reduces 160mv, and CF4 discharge amount reduces 0.39kg, and ton aluminium energy consumption reduces 1219.07KWh/t-Al.
S4: according to the resulting optimizing decision variable X of step S3bestIn control parameter come it is selected in rate-determining steps S2
Aluminium electrolytic industry scene, reaches the purpose of energy-saving and emission-reduction consumption reduction.
In above-described embodiment of the application, by providing a kind of intelligently control of the aluminium electroloysis energy-saving and emission-reduction based on A dominance relation
Method processed, firstly, being modeled using recurrent neural network to aluminum electrolysis process, then policymaker sets expectation target
Value introduces A-dominance preference administration method, in conjunction with multi-target quantum particle swarm algorithm (MQPSO) to production process model
It optimizes, is most met the desired optimizing decision variable of policymaker and corresponding current efficiency, tank voltage, perfluorinated
Object discharge amount and ton aluminium energy consumption.MQPSO algorithm does not need the complex operations such as to be intersected, made a variation, and only simplest position is more
New step, therefore cataloged procedure is simple, and introduces Quantum Properties, so that particle has strong ability of searching optimum, is easily guaranteed that
The integrality of the optimal value of preference, meets decisionmaker's preference demand during Evolution of Population.It is raw that aluminium electroloysis is obtained using this method
The optimal value of technological parameter during production can effectively improve current efficiency, reduce tank voltage, reduce greenhouse gas emissions and ton
Aluminium energy consumption, achieves energy-saving and emission reduction purposes.
Finally, it is stated that preferred embodiment above is only used to illustrate the technical scheme of the present invention and not to limit it, although logical
It crosses above preferred embodiment the present invention is described in detail, however, those skilled in the art should understand that, can be
Various changes are made to it in form and in details, without departing from claims of the present invention limited range.
Claims (7)
1. a kind of aluminium electroloysis modeling and optimization method based on recurrent neural network Yu angle preference, which is characterized in that including such as
Lower step:
S1: selection constitutes decision variable to current efficiency, tank voltage and the influential control parameter of perfluoro-compound discharge amount, certainly
Plan variable X=[x1,x2,···,xM], M is the number of selected control parameter;
S2: selected aluminium electrolytic industry scene acquires N group decision variable X1,X2,···,XNAnd its corresponding current efficiency y1,
y2,···,yN, tank voltage z1,z2,···,zNAnd perfluoro-compound discharge amount s1,s2,···,sNWith ton aluminium energy consumption
c1,c2,···,cNFor data sample, with each group of decision variable XiAs input, with corresponding current efficiency yi, tank voltage
ziAnd perfluoro-compound discharge amount siWith ton aluminium energy consumption ciAs output, sample is trained using recurrent neural network, is examined
It tests, establishes four aluminium cell production process models;
S3: using the preference multi-target quantum particle swarm algorithm dominated based on A, according to the preset desired value conduct of policymaker
Reference point, establishes the stringent partial ordering relation dominated based on A, and four production process models resulting to step S2 are optimized, obtained
Most meet the desired decision variable X of policymaker to one groupbestAnd its corresponding current efficiency ybest, tank voltage zbestAnd perfluor
Compound discharge amount sbestWith ton aluminium energy consumption cbest;
S4: according to the resulting optimizing decision variable X of step S3bestIn control parameter carry out selected aluminium electricity in rate-determining steps S2
Industry spot is solved, the purpose of energy-saving and emission-reduction consumption reduction is reached.
2. the aluminium electroloysis modeling and optimization method according to claim 1 based on recurrent neural network Yu angle preference,
It is characterized in that, in step S1, the control parameter includes potline current, blanking number, molecular proportion, aluminum yield, aluminum water is flat, is electrolysed
Matter level, bath temperature.
3. the aluminium electroloysis modeling and optimization method according to claim 1 based on recurrent neural network Yu angle preference,
It is characterized in that, in step S2, using current efficiency as output, establishes aluminium cell production process model, input layer uses 10
A neuron node, hidden layer use 15 neuron nodes, and output layer uses 1 neuron node, input layer to hidden layer
Between transmission function be Tansig function, hidden layer to the function between output layer is Purelin function, when sample training repeatedly
Generation number is 1000.
4. the aluminium electroloysis modeling and optimization method according to claim 1 based on recurrent neural network Yu angle preference,
It is characterized in that, in step S2, using tank voltage as output, establishes aluminium cell production process model, input layer uses 10
Neuron node, hidden layer use 15 neuron nodes, output layer use 1 neuron node, input layer to hidden layer it
Between transmission function be Logsig function, hidden layer to the function between output layer is Purelin function, iteration when sample training
Number is 1000.
5. the aluminium electroloysis modeling and optimization method according to claim 1 based on recurrent neural network Yu angle preference,
It is characterized in that, in step S2, using perfluoro-compound discharge amount as output, establishes aluminium cell production process model, input layer
Using 10 neuron nodes, hidden layer uses 15 neuron nodes, and output layer uses 1 neuron node, and input layer arrives
Transmission function is Logsig function between hidden layer, and hidden layer to the function between output layer is Purelin function, sample training
When the number of iterations be 1000.
6. the aluminium electroloysis modeling and optimization method according to claim 1 based on recurrent neural network Yu angle preference,
It is characterized in that, in step S2, using ton aluminium energy consumption as output, establishes aluminium cell production process model, input layer uses 10
A neuron node, hidden layer use 15 neuron nodes, and output layer uses 1 neuron node, input layer to hidden layer
Between transmission function be Tansig function, hidden layer to the function between output layer is Purelin function, when sample training repeatedly
Generation number is 1000.
7. the aluminium electroloysis modeling and optimization method according to claim 1 based on recurrent neural network Yu angle preference,
Be characterized in that, step S3 the following steps are included:
S31: the preference relation dominated according to A evaluates the fitness of each particle, and according to superiority and inferiority to individual optimal value and the overall situation
Optimal value is replaced;
S311: initialization system parameter, including population scale R, maximum number of iterations T generate n particle x at random1,
x2,···,xn, enable external archival collection Q for sky;
S312: policymaker sets preference angle [alpha] and preference intended reference point r (yp,zp,sp,cp), the preference intended reference point packet
Include the desired value of current efficiency, tank voltage, perfluoro-compound discharge amount and ton aluminium four targets of energy consumption;
S313: for each individual x, its fitness and its angle with reference point reference line are calculated:
Wherein, fjIt (x) is fitness value of the individual x in jth target;
S314: being based on angle information on object space, divides preference zone, if θ (r, x) < α, which is in preference
Region;Otherwise it is in non-preference zone;
S315: judge any two individual xiWith xkBetween superiority and inferiority relationship, including following situations:
Work as xiWith xkWhen being in preference zone or non-preference zone simultaneously, if xiPareto dominates xk, then it is assumed that xiIt is more excellent, if phase
Mutually Pareto is not dominated, then it is assumed that the two is suitable;
Work as xiIn preference zone, xkWhen in non-preference zone, if xiPareto dominates xkOr xiWith xkMutual not Pareto branch
Match, then it is assumed that xiBetter than xk, i.e. xiA dominates xk;
S316: the individual history optimal location pbest of each particle is determinedi, in system initialization, individual history optimal location
It is set as the initial position x of the particlei;After next iteration, based on the A dominance relation that S315 is proposed, to the new position x of particlei
With pbestiSuperiority and inferiority comparison is carried out, outstanding person saves as pbesti;
S317: updating external archival collection Q, is added to the particle that non-A is dominated between each other in population and achieves collection Q, and deletion is dominated
Particle;
S318: a particle is randomly choosed in external archival collection Q using press mechanism and Tabu search algorithm as global optimum position
It sets;
S32: Population Regeneration:
S321: the position of more new particle itself, wherein particle position more new formula are as follows:
Wherein: i (i=1,2 ..., n) represents i-th of particle, and n is population scale;J (j=1,2 ..., M) represents the jth of particle
Dimension, M are search space dimension;T is evolutionary generation;And uijIt (t) is equally distributed random number in [0,1] section;xij
(t),pbestij(t) and γij(t) respectively indicate the current location of particle i when evolutionary generation is t, individual history optimal location and
Attractor position;gbestj(t) and mbest (t) respectively indicates global optimum position and average best position when evolutionary generation is t
It sets;α indicates expansion-contraction factor;
S322: judging whether current globally optimal solution meets condition or whether the number of iterations reaches maximum number of iterations T, if
It is then to export current globally optimal solution, otherwise, the S321 that gos to step is computed repeatedly, until current globally optimal solution is full
Sufficient condition or the number of iterations reach maximum number of iterations T.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810193126.3A CN108984813B (en) | 2018-03-09 | 2018-03-09 | Aluminum electrolysis modeling and optimizing method based on recurrent neural network and angle preference |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810193126.3A CN108984813B (en) | 2018-03-09 | 2018-03-09 | Aluminum electrolysis modeling and optimizing method based on recurrent neural network and angle preference |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108984813A true CN108984813A (en) | 2018-12-11 |
CN108984813B CN108984813B (en) | 2022-12-13 |
Family
ID=64541753
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810193126.3A Active CN108984813B (en) | 2018-03-09 | 2018-03-09 | Aluminum electrolysis modeling and optimizing method based on recurrent neural network and angle preference |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108984813B (en) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2002070504A (en) * | 2000-09-05 | 2002-03-08 | Honda Motor Co Ltd | Blade shape designing method and information medium |
CN103903072A (en) * | 2014-04-17 | 2014-07-02 | 中国矿业大学 | High-dimensional multi-target set evolutionary optimization method based on preference of decision maker |
CN105207573A (en) * | 2015-09-04 | 2015-12-30 | 浙江大学 | Quantitative optimal configuration method of wind-solar hybrid power system based on discrete probability model |
CN105404926A (en) * | 2015-11-06 | 2016-03-16 | 重庆科技学院 | Aluminum electrolytic production technology optimization method based on BP neural network and MBFO algorithm |
CN105447567A (en) * | 2015-11-06 | 2016-03-30 | 重庆科技学院 | BP neural network and MPSO algorithm-based aluminium electrolysis energy-saving and emission-reduction control method |
CN106886656A (en) * | 2017-03-15 | 2017-06-23 | 南京航空航天大学 | A kind of cubical array antenna radiation pattern side lobe suppression method based on improvement MOPSO and convex optimized algorithm |
-
2018
- 2018-03-09 CN CN201810193126.3A patent/CN108984813B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2002070504A (en) * | 2000-09-05 | 2002-03-08 | Honda Motor Co Ltd | Blade shape designing method and information medium |
CN103903072A (en) * | 2014-04-17 | 2014-07-02 | 中国矿业大学 | High-dimensional multi-target set evolutionary optimization method based on preference of decision maker |
CN105207573A (en) * | 2015-09-04 | 2015-12-30 | 浙江大学 | Quantitative optimal configuration method of wind-solar hybrid power system based on discrete probability model |
CN105404926A (en) * | 2015-11-06 | 2016-03-16 | 重庆科技学院 | Aluminum electrolytic production technology optimization method based on BP neural network and MBFO algorithm |
CN105447567A (en) * | 2015-11-06 | 2016-03-30 | 重庆科技学院 | BP neural network and MPSO algorithm-based aluminium electrolysis energy-saving and emission-reduction control method |
CN106886656A (en) * | 2017-03-15 | 2017-06-23 | 南京航空航天大学 | A kind of cubical array antenna radiation pattern side lobe suppression method based on improvement MOPSO and convex optimized algorithm |
Non-Patent Citations (3)
Title |
---|
仲于江: "《基于小生境粒子群算法的柔性作业车间调度优化方法》", 《计算机集成制造系统》 * |
牛海帆: "《莱维飞行与粒子群的混合搜索算法》", 《太原科技大学学报》 * |
辜小花: "《基于演化模型偏好多目标优化的智能采油辅助决策支持》", 《机械工程学报》 * |
Also Published As
Publication number | Publication date |
---|---|
CN108984813B (en) | 2022-12-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109085752A (en) | Aluminium electroloysis preference multi-objective optimization algorithm based on angle dominance relation | |
CN108846526A (en) | A kind of CO2 emissions prediction technique | |
CN105447567B (en) | Aluminium electroloysis energy-saving and emission-reduction control method based on BP neural network Yu MPSO algorithms | |
CN105302973A (en) | MOEA/D algorithm based aluminum electrolysis production optimization method | |
CN105404926B (en) | Aluminum electrolysis production technique optimization method based on BP neural network Yu MBFO algorithms | |
CN109146121A (en) | The power predicating method stopped in the case of limited production based on PSO-BP model | |
CN105321000B (en) | Aluminum electrolysis process parameter optimization method based on BP neural network Yu MOBFOA algorithms | |
CN109242188B (en) | Long-term interval prediction and structure learning method for steel gas system | |
CN115470704B (en) | Dynamic multi-objective optimization method, device, equipment and computer readable medium | |
CN109670625A (en) | NOx emission concentration prediction method based on Unscented kalman filtering least square method supporting vector machine | |
CN114565239B (en) | Comprehensive low-carbon energy scheduling method and system for industrial park | |
CN108445756A (en) | Aluminium electroloysis energy-saving and emission-reduction intelligent control method based on AR dominance relations | |
CN105404142B (en) | Aluminium electroloysis multi parameters control method based on BP neural network Yu MBFO algorithms | |
CN111242270A (en) | Time series prediction model based on improved multi-target difference optimization echo state network | |
CN105426959B (en) | Aluminium electroloysis energy-saving and emission-reduction method based on BP neural network Yu adaptive M BFO algorithms | |
CN105334824A (en) | Aluminum electrolysis production optimization method based on NSGA-II algorithm | |
CN105302976A (en) | Aluminum electrolysis production optimization method based on SPEA2 algorithm | |
CN109086469A (en) | Aluminium electroloysis modeling and optimization method based on recurrent neural network and preference information | |
CN111626539A (en) | Power grid operation section dynamic generation method based on Q reinforcement learning | |
Li et al. | Prediction of grain yield in Henan Province based on grey BP neural network model | |
Fan et al. | Online learning-empowered smart management for A2O process in sewage treatment processes | |
CN109100995A (en) | Aluminium electroloysis energy-saving and emission-reduction optimization method based on decisionmaker's preference information | |
CN108984813A (en) | Aluminium electroloysis modeling and optimization method based on recurrent neural network Yu angle preference | |
CN108363303A (en) | Differential evolution aluminium electroloysis Multipurpose Optimal Method based on AR preference informations | |
CN105420760A (en) | Aluminum electrolysis production process multi-objective optimization method based on adaptive-step bacterial foraging algorithm |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |