CN108805346A - A kind of hot continuous rolling force forecasting method based on more hidden layer extreme learning machines - Google Patents
A kind of hot continuous rolling force forecasting method based on more hidden layer extreme learning machines Download PDFInfo
- Publication number
- CN108805346A CN108805346A CN201810565546.XA CN201810565546A CN108805346A CN 108805346 A CN108805346 A CN 108805346A CN 201810565546 A CN201810565546 A CN 201810565546A CN 108805346 A CN108805346 A CN 108805346A
- Authority
- CN
- China
- Prior art keywords
- hidden layer
- extreme learning
- learning machine
- continuous rolling
- optimal
- 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
- 238000005096 rolling process Methods 0.000 title claims abstract description 63
- 238000013277 forecasting method Methods 0.000 title description 8
- 238000013528 artificial neural network Methods 0.000 claims abstract description 73
- 239000002245 particle Substances 0.000 claims abstract description 30
- 238000000034 method Methods 0.000 claims abstract description 29
- 238000005457 optimization Methods 0.000 claims abstract description 24
- 230000002068 genetic effect Effects 0.000 claims abstract description 22
- 239000011159 matrix material Substances 0.000 claims description 32
- 230000006870 function Effects 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 6
- 239000010410 layer Substances 0.000 description 138
- 238000012549 training Methods 0.000 description 8
- 210000000349 chromosome Anatomy 0.000 description 6
- 238000012360 testing method Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 241001269238 Data Species 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000013178 mathematical model Methods 0.000 description 3
- 230000035772 mutation Effects 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 238000012795 verification Methods 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 244000144992 flock Species 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 210000004218 nerve net Anatomy 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000009785 tube rolling Methods 0.000 description 2
- 241000251468 Actinopterygii Species 0.000 description 1
- 229910002056 binary alloy Inorganic materials 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000019637 foraging behavior Effects 0.000 description 1
- 230000008303 genetic mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000002028 premature Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 239000002356 single layer Substances 0.000 description 1
- 238000010186 staining Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B21—MECHANICAL METAL-WORKING WITHOUT ESSENTIALLY REMOVING MATERIAL; PUNCHING METAL
- B21B—ROLLING OF METAL
- B21B38/00—Methods or devices for measuring, detecting or monitoring specially adapted for metal-rolling mills, e.g. position detection, inspection of the product
- B21B38/08—Methods or devices for measuring, detecting or monitoring specially adapted for metal-rolling mills, e.g. position detection, inspection of the product for measuring roll-force
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/04—Manufacturing
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Marketing (AREA)
- Biophysics (AREA)
- General Business, Economics & Management (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Tourism & Hospitality (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Mechanical Engineering (AREA)
- Manufacturing & Machinery (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The method for the hot continuous rolling power forecast based on more hidden layer extreme learning machine neural networks that the present invention relates to a kind of, includes the following steps:Obtaining influences the influence factor data of hot continuous rolling power numerical value;The influence factor data of acquisition are inputted into more hidden layer extreme learning machine prediction models;More hidden layer extreme learning machine prediction models export the prediction numerical value of hot continuous rolling power according to the influence factor data of input.Using multigroup existing hot continuous rolling power integrated data, genetic algorithm and particle swarm optimization algorithm, more hidden layer extreme learning machine prediction models based on more hidden layer extreme learning machine neural networks are established.A kind of forecasting procedure precision of prediction of more hidden layer extreme learning machine hot continuous rolling power provided by the invention is high, model easy care, while avoiding the drawback of the implicit number of plies and hidden layer node number that neural network is set according to designer personal experience, and precision is higher.
Description
Technical field
The invention belongs to field of computer technology more particularly to a kind of hot continuous rollings based on more hidden layer extreme learning machines
Force forecasting method.
Background technology
It is more next to the demand of Strip with the continuous prosperity and development of national economy and the continuous propulsion of modernization
More, this just puts forward higher requirements the production technology of Strip.In hot continuous rolling production process, strip reality is influenced
The principal element for shutting out thickness is roll-force, tube rolling simulation it is accurate whether be the key that realize accurate On-line Control.Tradition
Tube rolling simulation carried out by mathematical model, since process of establishing of mathematical model itself is ignored as and simplifies many
Actual field factor, thus it is larger by mathematical model calculating roll-force error merely, increasingly accurate rolling requirements cannot be met.
Drawbacks described above is that those skilled in the art it is expected to overcome
Invention content
(1) technical problems to be solved
For existing technical problem, the present invention is provided one kind and is rolled based on more hidden layer extreme learning machines forecast hot continuous rolling
The method of power processed.The method precision of prediction is high, model easy care, while avoiding rule of thumb setting Parameters of Neural Network Structure,
Precision is high.
(2) technical solution
In order to achieve the above object, the main technical schemes that the present invention uses are as follows:
A method of it forecasting hot continuous rolling power based on more hidden layer extreme learning machines, includes the following steps:
Obtaining influences the influence factor data of hot continuous rolling power numerical value;
The influence factor data of acquisition are inputted into more hidden layer extreme learning machine prediction models;
More hidden layer extreme learning machine prediction models export hot continuous rolling power according to the influence factor data of input
Predict numerical value.
Preferably, before the influence factor data of acquisition being inputted more hidden layer extreme learning machine prediction models, the method
Further include:
It is based on using multigroup existing hot continuous rolling power integrated data, genetic algorithm and particle swarm optimization algorithm, foundation
More hidden layer extreme learning machine prediction models of more hidden layer extreme learning machine neural networks.
Preferably, the method further includes following sub-step:
Obtain multigroup existing hot continuous rolling power integrated data, wherein wrapped in every group of hot continuous rolling power integrated data
N hot continuous rolling power influence factor is included, needs the hot continuous rolling power result forecast there are m, m, n are the integer more than 1;
The node parameter of more hidden layer limit learning neural network input and output layers is set, wherein the node of input layer is arranged
Number is n, and the node of setting multilayer extreme learning machine output layer is m;
Using the hot continuous rolling power influence factor as the input data of more hidden layer extreme learning machine neural networks,
Need the hot continuous rolling power result predicted as the output data of more hidden layer extreme learning machine neural networks using described;
The optimal implicit number of plies of more hidden layer extreme learning machine neural networks and optimal hidden layer knot are determined using genetic algorithm
Points;
The optimal weights matrix of the more hidden layer extreme learning machine neural networks obtained using particle swarm optimization algorithm and
Optimal bias vector;
The determining optimal implicit number of plies is set as to the implicit number of plies of more hidden layer extreme learning machine neural networks;
Set determining optimal hidden layer node number to the hidden layer node number of more hidden layer limit neural networks;
By the weight matrix that obtained optimal weights arranged in matrix is more hidden layer extreme learning machine neural networks;
Set obtained optimal bias vector to the bias vector of more hidden layer extreme learning machine neural networks;
The more hidden layer extreme learning machine neural networks obtained are more hidden layer extreme learning machine prediction models.
Preferably, the optimal hidden layer number of plies of more hidden layer extreme learning machines and optimal implicit is obtained using genetic algorithm
When layer nodal point number, first obtains optimal weights matrix of single hidden layer extreme learning machine under different hidden layer nodes and optimal be biased towards
Amount, then binary coding mode is used, according to the combination of the different implicit numbers of plies and different hidden layer node numbers, calculates prediction and miss
Difference selects the optimal implicit number of plies and optimal hidden layer node number according to prediction error.
Preferably, in the optimal weights for obtaining more hidden layer extreme learning machine neural networks using particle swarm optimization algorithm
When matrix and optimal bias vector, in each round iteration, particle after update, reinitializes every time according to the probability of setting
Particle.
Preferably, more hidden layer extreme learning machine neural networks are used as activation using hyperbolic tangent functions
Function.
(3) advantageous effect
A kind of hot continuous rolling force forecasting method based on more hidden layer extreme learning machines provided by the invention, to more hidden layer poles
After limit study machine neural network is using genetic algorithm and particle swarm optimization algorithm optimization, there is the optimal hidden layer number of plies and optimal hidden
The extreme learning machine neural network of nodal point number containing layer and optimal weights matrix and optimal bias vector is to influencing hot continuous rolling
The data of power predictors are handled, and the forecast result of roll-force is obtained, and avoid designer's sense datum setting nerve net
Network structural parameters, the precision of prediction finally obtained are high.
Description of the drawings
Fig. 1 is a kind of flow signal of hot continuous rolling force forecasting method based on more hidden layer extreme learning machines of the present invention
Figure;
Fig. 2 is two hidden layer ELM algorithm flow schematic diagrames.
Fig. 3 is to use genetic algorithm optimization ELM structural schematic diagrams.
Fig. 4 is the pre- to hot continuous rolling power based on the data for influencing hot continuous rolling power predictors of the embodiment of the present invention
The method schematic diagram of forecasting model is established in the method for report.
Fig. 5 is the reality of the hot continuous rolling power forecast data and hot continuous rolling power of test data in one embodiment of the invention
Border provides the scatter plot distributions of data.
Specific implementation mode
In order to preferably explain the present invention, in order to understand, below in conjunction with the accompanying drawings, by specific implementation mode, to this hair
It is bright to be described in detail.
As shown in Figure 1:Present embodiment discloses a kind of hot continuous rolling power forecast sides based on more hidden layer extreme learning machines
Method includes the following steps:
Obtaining influences the influence factor data of hot continuous rolling power numerical value;
The influence factor data of acquisition are inputted into more hidden layer extreme learning machine prediction models;
More hidden layer extreme learning machine prediction models export hot continuous rolling power according to the influence factor data of input
Predict numerical value.
A kind of hot continuous rolling force forecasting method based on more hidden layer extreme learning machines disclosed in the embodiment of the present invention leads to
Cross using established more hidden layer extreme learning machine prediction models to influence hot continuous rolling power numerical value influence factor data into
Row processing, obtains the result value of the hot continuous rolling power corresponding to influence factor data.
Wherein, more hidden layer extreme learning machine prediction models described here are to use multigroup existing hot continuous rolling in advance
Power integrated data, genetic algorithm and particle swarm optimization algorithm establish more hidden layers based on more hidden layer extreme learning machine neural networks
Extreme learning machine prediction model.
There is further provided the method for building up of more hidden layer extreme learning machine prediction models described in the above method, specifically include
Following steps:
Obtain multigroup existing hot continuous rolling power integrated data, wherein wrapped in every group of hot continuous rolling power integrated data
N hot continuous rolling power influence factor is included, needs the hot continuous rolling power result forecast there are m, m, n are the integer more than 1;
The node parameter of more hidden layer limit learning neural network input and output layers is set, wherein the node of input layer is arranged
Number is n, and the node of setting multilayer extreme learning machine output layer is m;
Using the hot continuous rolling power influence factor as the input data of more hidden layer extreme learning machine neural networks,
Need the hot continuous rolling power result predicted as the output data of more hidden layer extreme learning machine neural networks using described;
The optimal implicit number of plies of more hidden layer extreme learning machine neural networks and optimal hidden layer knot are determined using genetic algorithm
Points;
The optimal weights matrix of the more hidden layer extreme learning machine neural networks obtained using particle swarm optimization algorithm and
Optimal bias vector;
The determining optimal implicit number of plies is set as to the implicit number of plies of more hidden layer extreme learning machine neural networks;
Set determining optimal hidden layer node number to the hidden layer knot of more hidden layer extreme learning machine neural networks
Points;
By the weight matrix that obtained optimal weights arranged in matrix is more hidden layer extreme learning machine neural networks;
Set obtained optimal bias vector to the bias vector of more hidden layer extreme learning machine neural networks;
The more hidden layer extreme learning machine neural networks obtained are more hidden layer extreme learning machine prediction models.
The optimal of more hidden layer extreme learning machine neural networks is obtained in the present embodiment using genetic algorithm to imply layer by layer
Number and when optimal hidden layer node number, first obtain optimal weights matrix of single hidden layer extreme learning machine under different hidden layer nodes and
Optimal bias vector, then binary coding mode is used, according to the combination of the different implicit numbers of plies and different hidden layer node numbers, meter
Prediction error is calculated, according to prediction error, selects the optimal implicit number of plies and optimal hidden layer node number.
It is further to note that:More hidden layer extreme learning machine neural networks are being obtained using particle swarm optimization algorithm
Optimal weights matrix and when optimal bias vector, in each round iteration, particle is every time after update, according to the probability of setting
Reinitialize particle.
More hidden layer extreme learning machine neural networks described in the present embodiment use hyperbolic tangent function conducts
Activation primitive.
But the roll-force based on more hidden layer extreme learning machine neural networks can established by obtaining multigroup data to be predicted
Before the step of forecasting model, or later.
Specifically, the integrated data for obtaining multigroup known roll-force is being established based on more hidden layer extreme learning machine neural networks
Before the step of hot continuous rolling power forecasting model, it is adapted to the modelling phase.
Specifically, the integrated data of known roll-force is uniformly selected into a part as the training sample for establishing model
This, using remaining sample as the verification sample of verification model.
Specifically, multigroup data to be predicted are obtained and are establishing the hot continuous rolling based on more hidden layer extreme learning machine neural networks
After the step of draught pressure forecast model, it is adapted to after model foundation carry out hot continuous rolling power forecast rank using model
Section.
Specifically, after establishing based on the hot continuous rolling power forecasting model of more hidden layer extreme learning machine neural networks,
The data that forecast Re Lianzha roll-forces are needed to acquisition, handle these using more hidden layer extreme learning machine prediction models of foundation and wait for
Forecast data obtains the roll-force numerical value of prediction.
By training step above-mentioned and verification step, forecasts that the accuracy rate of roll-force is higher, produce reality can be met
Needs.
Specifically, extreme learning machine (Extreme Learning Machine, hereinafter referred to as ELM) neural network is one
The feedforward neural network of single hidden layer the advantage is that network training speed is very fast, and fitting precision is high, and Generalization Capability is good.But
Be single layer extreme learning machine neural network network structure it is determined according to the practical experience of oneself by designer, this warp
The selection for the property tested is not most reasonable and optimal, needs to carry out test of many times in actual application, chooses optimal network mould
Type increases artificial amount, also takes, and the parameter of extreme learning machine is random initializtion, leads to part weights and deviation
Not up to optimum state.
And genetic algorithm (Genetic Algorithms, hereinafter referred to as GA) and particle group optimizing (Particle Swarm
Optimization, hereinafter referred to as PSO) algorithm is global optimization approach, therefore by genetic algorithm, extreme learning machine nerve net
Network and particle swarm optimization algorithm combine, and overcome their respective disadvantages, its advantage are made full use of, using genetic algorithm to more
The hidden layer number of plies and hidden layer node number of hidden layer extreme learning machine carry out global optimization, are learnt to the limit using particle cluster algorithm
The weight matrix and bias vector of machine neural network carry out global optimization, and obtaining has the optimal hidden layer number of plies and optimal hidden layer
More hidden layer extreme learning machine neural networks of nodal point number, optimal weights matrix and optimal bias matrix can be used for forecasting that Re Lianzha rolls
Power processed.
Extreme learning machine neural network
For the sample (x that sample size is Ni, ti), wherein xi=[xi1, xi2, xi3..., xin]T∈Rn,
ti=[ti1, ti2..., tim]T∈Rm, then the Single hidden layer feedforward neural networks of L hidden layer neuron of standard
For:
Here, ai=[ai1, ai2..., ain]T∈RnBetween expression i-th of neuron of input layer and hidden layer
Input weights;βiTo export weights, biFor biasing;ai·xjIndicate aiAnd xjInner product.Then the object function of ELM can be another
Kind mode is expressed as:
H β=T (2)
Here:
β=[β1..., βL]T, T=[t1..., tL]T (4)
H is the matrix of the hidden layer output of neural network;T is desired output.The algorithm of Huang professors is that random selection is defeated
Enter weights and hidden layer deviation, this network structure of training is equivalent to the least square solution for solving linear system H β=T
Minimize:||Hβ-T|| (5)
The minimum value of the last professor Huang verified least-square solution of linear system is:
Here:H+It is the Moore-Penrose generalized inverse matrix of H, and the solution of H β=T is unique.
More hidden layer extreme learning machine neural networks
As shown in Figure 2:For the sample (x that sample size is Ni, ti), wherein
xi=[xi1, xi2, xi3..., xin]T∈Rn, ti=[ti1, ti2..., tim]T∈Rm, and all hidden layers all have
There are identical hidden layer node number and activation primitive, is brief Jie by taking double hidden layer ELM (TELM) neural networks as an example herein
It continues, its two hidden layers is first regarded as the ELM containing single hidden layer, the output that can thus find out hidden layer is
H=WX+B (7)
H is hidden layer output matrix, and W is the weight matrix of input layer and hidden layer, and B is bias matrix, can be asked by formula (6)
Go out the weight matrix β between hidden layer and output layer, second hidden layer is added in ELM neural networks now, restores two
The neural network structure of a hidden layer, the prediction output that we can obtain second hidden layer are
H1=g (W1H+B1) (8)
W1For the weight matrix between first hidden layer and second hidden layer, B1For being biased towards for second hidden layer
Amount, and the desired output of second hidden layer is H1*=T β+ (9)
It is infinite close to desired output to meet prediction output so that H1*=H1If WHE=[B1 W1], you can it acquires
For matrix HE=[1 H]TGeneralized inverse matrix, 1 indicates a vector for having Q element, and each element
All it is 1, g-1(x) be activation primitive g (x) inverse function.
After the weight matrix and bias vector of second hidden layer all solve, the pre- of second hidden layer may be updated
Surveying output is:
H2=g (w1h+B1)=g (WHEHE) (11)
Therefore, hidden layer output matrix β may be updated as:
Final neural network output f (x), which can be obtained, is
F (x)=H2βnew (13)
Genetic algorithm
Genetic algorithm is a kind of made of the simulation nature genetic mechanism and theory of biological evolution proposed by professor Holland
Parallel stochastic optimization method, the coding mode of chromosome include mainly binary system hair, real number method etc., it mainly by selecting,
Intersect and individual is screened in variation, retains the good individual of fitness, GA algorithms have efficient heuristic search, parallel to count
Calculate, it is easy to implement, it is applied widely the advantages that, so being obtained in optimization problem commonly used.The realization of genetic algorithm
Journey is as follows:Each chromosome is seen as a particle, and selection operation refers to the fitness value by calculating different chromosomes, root
According to roulette method, from old group in certain probability selection individual to new group;Crossover operation refers to that two are selected from group
A chromosome, random selection one-point or multi-point chromosome location swap;Mutation operation refer to from group optionally one by one
Body has a little generated more excellent individual into row variation in selective staining body.In existing research, the hidden layer knot of ELM
Points be all according to the empirically determined of designer, it is not necessarily optimal, in the present invention, using the binary coding of genetic algorithm
The hidden layer number of plies and hidden layer node number of more hidden layer extreme learning machine neural networks is optimized in mode, specifically, institute
It states in method, in the optimal hidden layer number of plies for obtaining more hidden layer extreme learning machine neural networks using genetic algorithm and implies
When layer nodal point number, the implicit number of plies and hidden layer node number are first randomly generated, more hidden layer extreme learning machine neural networks are built, calculated
The fitness value of individual obtains the optimal hidden layer number of plies and optimal hidden layer node by selection, intersection, mutation operation
Several combinations.
Particle swarm optimization algorithm
Population is a kind of global optimization approach proposed by Kennedy and Eberhart, and research idea, which inspires, comes from fish
Group and flock of birds foraging behavior.Since PSO algorithm structures are simple, easily realize, it is powerful etc., so being obtained in optimization problem
It is commonly used.Particle swarm optimization algorithm experimentation is as follows:In a flock of birds, each bird is seen as a particle, they
There is corresponding speed Vi=(Vi1, Vi2..., Vid) and position Xi=(XI1, Xi2..., Xid) wherein i=1,2,3 ..., m.Every
In a generation, population constantly updates position by following formula (14) and (15), to find out optimal location, and records.
Here, pBesti=(pBesti1, pBesti2..., pBestid) it is the optimal location that particle itself is undergone;
gBesti=(gBesti1, gBesti2..., gBestid) it is the optimal location passed through in all populations, i=1,2 ..., m;K is
Current iteration number;W is inertia weight;c1, c2It is accelerator coefficient;r1, r2It is to be uniformly distributed random function on [0,1].It is existing
The shortcomings of some PSO algorithms, there is Premature Convergences, and search precision is relatively low, and later stage iteration is inefficient, uses for reference heredity and calculates here
The variation thought of method, mutation operation is introduced in PSO algorithms, after particle updates every time, is reinitialized with certain probability
Particle.
Specifically, 134 groups of data are obtained, wherein choosing 100 groups is used as training data, 34 groups are used as test data.Using
100 groups of training datas, which are established, has the optimal implicit number of plies and hidden layer node number, optimal weights matrix and optimal bias vector
More hidden layer extreme learning machine neural network prediction models.Such as Fig. 3, shown in Fig. 4, the embodiment of the present invention based on more hidden layer limit
The method that more hidden layer extreme learning machine prediction models are established in the hot continuous rolling force forecasting method of habit machine.
Steps are as follows:
(1) initial time genetic algorithm, it is 100 that GA, which selects iterations, chromosome quantitative 20, crossover probability 0.4, is become
Different probability is 0.2, code length 9;Select hyperbolic tangent functions as ELM neural network activation primitives.
(2) it will forecast that the data of influence factor input to ELM neural networks comprising hot continuous rolling power, use ELM algorithms
Data training to the influence factor forecast including hot continuous rolling power, verifies the fitness value of GA, sentences to fitness value
It is disconnected, then preserve optimal value;
(3) optimizing is just exited when network reaches maximum iteration, more hidden layer extreme learning machines at this moment just have most
The excellent implicit number of plies and optimal hidden layer node number;
(4) PSO is initialized, it is 200 to select iterations, and population quantity is 30;
(5) using having determined more hidden layer ELM neural networks of optimum structure to the shadow that is forecast comprising hot continuous rolling power
The data training of the factor of sound, verifies the fitness value of PSO, judges fitness value, according to formula (14) (15), to particle
Speed and position are updated, and according to the probability of setting, reinitialize some particles, then preserve the speed of optimal value and PSO
Degree and position.
(6) optimizing is exited when network reaches maximum iteration, more hidden layer extreme learning machine neural networks tool at this moment
There are optimal weights matrix and optimal bias vector.
34 groups of test datas are handled using the hot continuous rolling power forecasting model, obtained forecast result is as shown in Figure 5.
As shown in figure 5, handled in 34 groups of test datas using the forecasting model, the predicted value of every group of roll-force all with set
Definite value has reached preferable fitting.
In conclusion compared with the method for tradition forecast hot continuous rolling power, the embodiment of the present invention is based on more hidden layer limit
Learn the forecasting procedure of machine neural network, the precise and high efficiency when forecasting hot continuous rolling power.The embodiment of the present invention based on how hidden
The hot continuous rolling force forecasting method of layer extreme learning machine neural network, can in speed, accuracy have incomparable advantage and
Important actual application value.
Finally it should be noted that above-described each embodiment is only limitted to illustrate technical scheme of the present invention, rather than to it
Limitation;Although the present invention is described in detail referring to the foregoing embodiments, it will be understood by those of ordinary skill in the art that;
Its technical solution that can be still recorded to previous embodiment is modified, or is carried out to which part or all technical features
Equivalent replacement;And these modifications or substitutions, the essence of corresponding technical solution can't be made to be detached from the present invention embodiment technical side
The range of case.
Claims (6)
1. a kind of method for forecasting hot continuous rolling power based on more hidden layer extreme learning machines, which is characterized in that include the following steps:
Obtaining influences the influence factor data of hot continuous rolling power numerical value;
The influence factor data of acquisition are inputted into more hidden layer extreme learning machine prediction models;
More hidden layer extreme learning machine prediction models export the prediction of hot continuous rolling power according to the influence factor data of input
Numerical value.
2. the method as described in claim 1, which is characterized in that the influence factor data of acquisition are inputted more hidden layer limit and are learnt
Before machine prediction model, the method further includes:
Using multigroup existing hot continuous rolling power integrated data, genetic algorithm and particle swarm optimization algorithm, establish based on how hidden
More hidden layer extreme learning machine prediction models of layer extreme learning machine neural network.
3. method as claimed in claim 2, which is characterized in that the method further includes following sub-step:
Obtain multigroup existing hot continuous rolling power integrated data, wherein every group of hot continuous rolling power integrated data includes n
Hot continuous rolling power influence factor needs the hot continuous rolling power result forecast to have m, and m, n are the integer more than 1;
The node parameter of more hidden layer limit learning neural network input and output layers is set, wherein the node number of input layer is arranged
Node for n, setting multilayer extreme learning machine output layer is m;
Using the hot continuous rolling power influence factor as the input data of more hidden layer extreme learning machine neural networks, by institute
State the output data for needing the hot continuous rolling power result predicted as more hidden layer extreme learning machine neural networks;
The optimal implicit number of plies of more hidden layer extreme learning machine neural networks and optimal hidden layer node number are determined using genetic algorithm;
The optimal weights matrix of the more hidden layer extreme learning machine neural networks obtained using particle swarm optimization algorithm and optimal
Bias vector;
The determining optimal implicit number of plies is set as to the implicit number of plies of more hidden layer extreme learning machine neural networks;
Set determining optimal hidden layer node number to the hidden layer node number of more hidden layer limit neural networks;
By the weight matrix that obtained optimal weights arranged in matrix is more hidden layer extreme learning machine neural networks;
Set obtained optimal bias vector to the bias vector of more hidden layer extreme learning machine neural networks;
The more hidden layer extreme learning machine neural networks obtained are more hidden layer extreme learning machine prediction models.
4. method as claimed in claim 3, which is characterized in that
When obtaining the optimal hidden layer number of plies of more hidden layer extreme learning machines and optimal hidden layer node number using genetic algorithm,
First obtain optimal weights matrix and optimal bias vector of single hidden layer extreme learning machine under different hidden layer nodes, then using two into
Coding mode processed calculates prediction error, is missed according to prediction according to the combination of the different implicit numbers of plies and different hidden layer node numbers
Difference selects the optimal implicit number of plies and optimal hidden layer node number.
5. method as claimed in claim 3, which is characterized in that
In the optimal weights matrix that obtains more hidden layer extreme learning machine neural networks using particle swarm optimization algorithm and optimal
When bias vector, in each round iteration, after update, particle is reinitialized according to the probability of setting every time for particle.
6. method as claimed in claim 3, which is characterized in that
More hidden layer extreme learning machine neural networks are using hyperbolic tangent functions as activation primitive.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810565546.XA CN108805346A (en) | 2018-06-04 | 2018-06-04 | A kind of hot continuous rolling force forecasting method based on more hidden layer extreme learning machines |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810565546.XA CN108805346A (en) | 2018-06-04 | 2018-06-04 | A kind of hot continuous rolling force forecasting method based on more hidden layer extreme learning machines |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108805346A true CN108805346A (en) | 2018-11-13 |
Family
ID=64088542
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810565546.XA Pending CN108805346A (en) | 2018-06-04 | 2018-06-04 | A kind of hot continuous rolling force forecasting method based on more hidden layer extreme learning machines |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108805346A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110349680A (en) * | 2019-06-27 | 2019-10-18 | 武汉科技大学 | A kind of prediction technique of opioid drug crisis explosion time |
CN111889524A (en) * | 2020-07-31 | 2020-11-06 | 中冶赛迪重庆信息技术有限公司 | Cold rolling force prediction method and system based on machine learning method |
CN112037209A (en) * | 2020-09-02 | 2020-12-04 | 上海大学 | Steel plate roller wear loss prediction method and system |
CN112487700A (en) * | 2020-09-15 | 2021-03-12 | 燕山大学 | Cold rolling force prediction method based on NSGA and FELM |
CN112653751A (en) * | 2020-12-18 | 2021-04-13 | 杭州电子科技大学 | Distributed intrusion detection method based on multilayer extreme learning machine in Internet of things environment |
WO2022116571A1 (en) * | 2020-12-04 | 2022-06-09 | 东北大学 | Lstm-based roll-bending force prediction method for hot rolling |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103400190A (en) * | 2013-08-13 | 2013-11-20 | 浙江大学 | Integrated framework method for optimizing extremity learning machine by using genetic algorithm |
CN104156560A (en) * | 2014-07-12 | 2014-11-19 | 中国矿业大学 | Multi-level coal mine water inrush prediction method based on SaE-ELM (self-adaptive evolutionary extreme learning machine) |
CN104200268A (en) * | 2014-09-03 | 2014-12-10 | 辽宁大学 | PSO (Particle Swarm Optimization) extremity learning machine based strip steel exit thickness predicting method |
-
2018
- 2018-06-04 CN CN201810565546.XA patent/CN108805346A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103400190A (en) * | 2013-08-13 | 2013-11-20 | 浙江大学 | Integrated framework method for optimizing extremity learning machine by using genetic algorithm |
CN104156560A (en) * | 2014-07-12 | 2014-11-19 | 中国矿业大学 | Multi-level coal mine water inrush prediction method based on SaE-ELM (self-adaptive evolutionary extreme learning machine) |
CN104200268A (en) * | 2014-09-03 | 2014-12-10 | 辽宁大学 | PSO (Particle Swarm Optimization) extremity learning machine based strip steel exit thickness predicting method |
Non-Patent Citations (3)
Title |
---|
DONG XIAO,ETC.: "A Multiple Hidden Layers Extreme Learning Machine Method and Its Application", 《MATHEMATICAL PROBLEMS IN ENGINEERING》 * |
孔祥松 等: "基于GA - BP 的煤巷围岩稳定性分类与应用", 《煤矿安全》 * |
曹卫华 等: "基于极限学习机的热轧薄板轧制力预测模型", 《信息与控制》 * |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110349680A (en) * | 2019-06-27 | 2019-10-18 | 武汉科技大学 | A kind of prediction technique of opioid drug crisis explosion time |
CN111889524A (en) * | 2020-07-31 | 2020-11-06 | 中冶赛迪重庆信息技术有限公司 | Cold rolling force prediction method and system based on machine learning method |
CN112037209A (en) * | 2020-09-02 | 2020-12-04 | 上海大学 | Steel plate roller wear loss prediction method and system |
CN112037209B (en) * | 2020-09-02 | 2023-07-28 | 上海大学 | Method and system for predicting abrasion loss of steel plate roller |
CN112487700A (en) * | 2020-09-15 | 2021-03-12 | 燕山大学 | Cold rolling force prediction method based on NSGA and FELM |
CN112487700B (en) * | 2020-09-15 | 2022-04-19 | 燕山大学 | Cold rolling force prediction method based on NSGA and FELM |
WO2022116571A1 (en) * | 2020-12-04 | 2022-06-09 | 东北大学 | Lstm-based roll-bending force prediction method for hot rolling |
CN112653751A (en) * | 2020-12-18 | 2021-04-13 | 杭州电子科技大学 | Distributed intrusion detection method based on multilayer extreme learning machine in Internet of things environment |
CN112653751B (en) * | 2020-12-18 | 2022-05-13 | 杭州电子科技大学 | Distributed intrusion detection method based on multilayer extreme learning machine in Internet of things environment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108805346A (en) | A kind of hot continuous rolling force forecasting method based on more hidden layer extreme learning machines | |
Nazari-Heris et al. | Short-term scheduling of hydro-based power plants considering application of heuristic algorithms: A comprehensive review | |
CN108133258B (en) | Hybrid global optimization method | |
CN110084424A (en) | A kind of Methods of electric load forecasting based on LSTM and LGBM | |
CN104636985B (en) | A kind of transmission line of electricity radio interference Forecasting Methodology of improved BP | |
CN108734202A (en) | A kind of Fault Diagnosis for HV Circuit Breakers method based on improved BP | |
CN110222883A (en) | Load Prediction In Power Systems method based on wind Drive Optimization BP neural network | |
CN108876001A (en) | A kind of Short-Term Load Forecasting Method based on twin support vector machines | |
CN104539601B (en) | Dynamic network attack process analysis method for reliability and system | |
CN110535149A (en) | A kind of area Gong Biantai electric load three-phase imbalance prediction technique | |
CN108204944A (en) | The Buried Pipeline rate prediction method of LSSVM based on APSO optimizations | |
CN105335619A (en) | Collaborative optimization method applicable to parameter back analysis of high calculation cost numerical calculation model | |
CN109214565A (en) | A kind of subregion system loading prediction technique suitable for the scheduling of bulk power grid subregion | |
CN108182490A (en) | A kind of short-term load forecasting method under big data environment | |
CN104732067A (en) | Industrial process modeling forecasting method oriented at flow object | |
CN102323949A (en) | Keyword optimization classification method based on fuzzy genetic algorithm | |
Jain et al. | A novel hybrid method for short term load forecasting using fuzzy logic and particle swarm optimization | |
CN113935556B (en) | Temperature sensor optimal arrangement method based on DNA genetic algorithm | |
Wu et al. | Short-Term Electric Load Forecasting Model based on PSO-BP | |
Niu et al. | Research on neural networks based on culture particle swarm optimization and its application in power load forecasting | |
CN113762591B (en) | Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning | |
CN115310348A (en) | Stacking-based grouting amount integrated agent prediction model and prediction method | |
Huo et al. | An improved artificial bee colony algorithm for numerical functions | |
CN114021847A (en) | Short-term power load prediction method and system based on ISGU (inverse transformation and aggregation) hybrid model | |
Han et al. | Improved FOA-ESN method using opposition-based learning mechanism for the network traffic prediction with multiple steps |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181113 |