CN106991208A - Forecasting Methodology based on the injector performance using the BP artificial neural networks for improving mind evolutionary - Google Patents
Forecasting Methodology based on the injector performance using the BP artificial neural networks for improving mind evolutionary Download PDFInfo
- Publication number
- CN106991208A CN106991208A CN201710111923.8A CN201710111923A CN106991208A CN 106991208 A CN106991208 A CN 106991208A CN 201710111923 A CN201710111923 A CN 201710111923A CN 106991208 A CN106991208 A CN 106991208A
- Authority
- CN
- China
- Prior art keywords
- sub
- individual
- population
- group
- injector
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/17—Mechanical parametric or variational design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
A kind of Forecasting Methodology of the injector performance based on using the BP artificial neural networks for improving mind evolutionary, relevant parameter is collected for given injector；Mind evolutionary initiation parameter is set；Several winning individuals and temporary individual of highest scoring in initial population are calculated according to scoring function, new individual is produced centered on these individuals, winning sub- population and interim sub- population is obtained；Operation similartaxis and operation dissimilation；The sub-group for keeping score of a relatively high in the relatively low sub-group of score progress cross and variation is obtained into new individual, remaining sub-group is eliminated again and re-started in global scope and searches for and formed new interim colony, the scoring highest individual of generation is required optimized individual；Neural network structure is set up and parameter initialization；Improve the training of BP MEA neutral nets；The measured data of the given injector of collection, is input in improvement BP MEA neutral nets, obtains output vector i.e. predicted value.Precision of prediction of the present invention is higher, time-consuming shorter.
Description
Technical field
It is especially a kind of to be based on using mind evolutionary (MEA) the present invention relates to a kind of injector performance Forecasting Methodology
BP artificial neural networks injector performance Forecasting Methodology.
Background technology
Injector produces vacuum, and the relatively low fluid of suction pressure by the higher fluid of one pressure by main jet,
The pressure of fluid is improved after mixing by diffuser, the fluid of middle pressure is finally given, i.e., by the lifting of low pressure fluid pressure,
Realize the effect of compression.Injector can by the use of the low-grade energies such as industrial overbottom pressure, waste heat, used heat, solar heat, underground heat as
Driving, without consumption electric power, with good effects of energy saving and emission reduction, under the energy environment background that the situation is tense, in chemical industry, heat
The field such as energy, refrigeration, HVAC is widely applied with largely paying close attention to and studying.
Mass ratio of induced-to-inducing air (ε) and outlet back pressure (p under critical excitationc) it is the crucial performance parameter of injector.But by
Flowed in injector interior it is extremely complex, including be jammed twice, Supersonic Flow, all kinds of shock waves, the phenomenon such as fan-shaped diffusion, use
The parameters precision that one-dimensional physical model simulation is obtained is relatively low, and effect is poor, such as (the W.Chen et al.Theoretical of document 1
analysis of ejector refrigeration system performance under overall
modes.Applied energy,185-2:Injection refrigerating system performance under the full working scopes such as 2074-2084,2016, i.e. W.Chen
The theory analysis application energy, 185-2:2074-2084,2016.) and (the JM.Cardemil et al.A of document 2
general model for evaluation of vapor ejectors performance for application in
refrigeration.Energy Conversion and Management,64:The such as 79-86,2012, i.e. JM.Cardemil
One model energy for being used for cooling steam jet ejector Performance Evaluation is changed and management, 64:79-86,2012.) shown in
The mean error of conventional model is more in 5-10%, and worst error is up to more than 15%.The method for such as using Fluid Mechanics Computation
Long, also labor intensive material resources are then taken, are not suitable for design and the research of associated cyclic.Design of the above-mentioned present situation to injector should
Problem is brought with work such as, associated cyclic researchs.
The content of the invention
In order to which the precision of prediction for the Forecasting Methodology for overcoming the shortcomings of existing injector performance is relatively low, time-consuming longer, the present invention
It is higher, time-consuming shorter based on using the BP artificial neural networks' for improving mind evolutionary there is provided a kind of precision of prediction
The Forecasting Methodology of injector performance.
The technical solution adopted for the present invention to solve the technical problems is：
A kind of Forecasting Methodology of the injector performance based on using the BP artificial neural networks for improving mind evolutionary,
Comprise the following steps：
Step one, the collection and processing of data：For given injector, it is citation jet body pressure to collect relevant parameter
Pe, working fluid pressure Pp, outlet back pressure Pc and mass ratio of induced-to-inducing air ε, to citation jet body pressure, working fluid pressure and outlet back pressure
It is normalized, it is arrived between [0,1], formula is as follows：
Wherein, k is the data after normalization, and x is to be normalized data, xminTo be normalized the minimum value in data,
xmaxTo be normalized maximum in data；
Step 2, mind evolutionary initiation parameter is set：To including Population Size, winning sub- population number, interim
The parameters such as sub- population number, sub-population size, iterations are configured and set up initial population；
Step 3, several winning individuals and temporary individual of highest scoring in initial population are calculated according to scoring function,
New individual is produced centered on these individuals, winning sub- population and interim sub- population is obtained；
Individual in step 4, operation similartaxis, each sub- population is vied each other, until ripe sub- population is produced, by score
Recorded；
Step 5, operation dissimilation, the optimum individual of each ripe sub- population carries out global competition and forms new interim group
Body；
Step 6, the sub-group for keeping score of a relatively high in the relatively low sub-group of score progress cross and variation is obtained newly
Individual, then remaining sub-group is eliminated and re-started in global scope search for and form new interim colony；
Step 7, four~step 6 of repeat step reaches the iterations upper limit, and the scoring highest individual now produced is i.e.
For required optimized individual；
Step 8, neural network structure is set up and parameter initialization：According to the input and output vector of network, nerve net is determined
Network structure, it is one layer to imply the number of plies, neuron number rule of thumb formula：Draw, wherein l, n for input,
Output node number, a is that constant takes 1~10；Wherein, citation jet body pressure, working fluid pressure are the input of the neutral net, are gone out
Mouth back pressure, the output that mass ratio of induced-to-inducing air is the neutral net；
Step 9, improves the training of BP-MEA neutral nets：Instructed with the artificial neural network set up in step (8)
Practice, by iteration several times, until predicted value and desired value are less than set-point, now neural metwork training terminates, and the training is complete
Into neutral net be foundation injector performance predict BP-MEA neutral nets；
Step 10, in practical implementation, the measured data of the given injector of collection, including citation jet body pressure Pe,
Working fluid pressure Pp；Data are normalized by the method for step one, the improvement set up and completed then is input to again
In BP-MEA neutral nets, output vector outlet back pressure Pc and mass ratio of induced-to-inducing air ε are obtained, then outlet back pressure is carried out at renormalization
Reason, that is, obtain predicted value.
Further, in the step 2, mind evolutionary parameter initialization, such as Population Size, winning sub- population number
And interim sub- population number, it is each individual to be then made up of the weights and threshold value of neutral net.This BP-MEA neural network structures
There is l node for l-m-n, i.e. input layer, hidden layer has m node, and output layer has n node, so shared l × m × n power
Value, m+n threshold value, so this algorithm individual UVR exposure length is and weights and threshold value are random number between [- 1,1]；Using upper
State coding and set up evaluation function, individual is scored, score the higher BP-MEA set up with the individual by weights and threshold value
Predicting the outcome for neutral net is more accurate.
Further, in the step 3, S individual is produced, is scored according to step 3 and according to score size liter
Sequence is arranged, and produces N1The individual winning individual of scoring highest and N2The higher temporary individual of individual scoring, centered on above-mentioned individual,
H new individuals are produced around each individual, so as to obtain the N that number of individuals is H1Individual winning sub- population and N2Individual interim sub- population.
Further, in the step 4, operation similartaxis process is：Individual in each sub- population is vied each other, and is chosen simultaneously
Record optimal solution is produced new victor, if no longer producing new victor, and the sub-group is ripe sub-group, convergent process
Ripe winning population and ripe interim population are formed after end, the score that optimum individual in each sub- population is recorded respectively is used as this
The score of sub- population.
In the step 5, operation dissimilation process is：The optimum individual of each ripe sub- population carries out global competition, if facing
When sub-group score of the score higher than some ripe winning sub-group, then the interim subgroup that the winning sub-group can be won
Body is substituted, and the individual in former winning sub-group is released；If the score of a ripe interim sub-group is excellent less than any one
Win the score of sub-group, then the interim sub-group is gone out of use, and individual therein is released.
In the step 6, the interim sub-group eliminated is discharged and scans for re-forming newly in global scope
Interim sub-group, choose and optimum individual showed in preferable two sub-groups in winning sub-group using the progress of real number interior extrapolation method
Crossover operation, by k-th of chromosome a in the colonykWith l-th of chromosome alIt is as follows in the crossover operation methods of j：
akj=akj(1-r)+aljr
alj=alj(1-r)+akjr
Wherein, r is the random number between [0,1], produces new individual and S new are produced using new individual as center
Body simultaneously carries out operation similartaxis and obtains carrying out operation dissimilation after ripe sub-group again.
In the step 7, the iterations upper limit is set X times, the scoring highest individual of generation is required optimal
Body.
In the step 8, the foundation of BP-MEA neural network structures and the initialization of relevant parameter are improved：By optimized individual
Comprising threshold value decoded with weights, build BP artificial neural networks, according to neutral net input sample, i.e. driving fluid
Pressure, working fluid pressure, export sample, i.e. outlet back pressure, mass ratio of induced-to-inducing air, between input layer, hidden layer and each layer of output layer
Connection weight initialization value take [- 1,1] at random, use ωij、ωjkRepresent, learning rate η takes 0.1~0.2, and training objective takes 10-3
~10-6。
In the step 9, input training sample starts to be trained process to neutral net as follows：
9.1) calculating of hidden layer：Wherein l, m, n represent input layer number, node in hidden layer and output layer section respectively
Points, f (x) is that transmission function takes S types (Sigmoid) function, and x is the data that output layer is inputted；
The input of hidden layer node
The output H of hidden layer nodej=f (Sj)
9.2) calculating of output layer：Wherein YbFor the prediction output during neural network；
Export the output of node layer
9.3) error calculation：K-th of the neuron prediction of neutral net output layer is output as Ybk, YkFor k-th neuron
, there is error e between them in desired outputk, formula is as follows：
ek=Yk-Ybk
9.4) renewal of weights：According to error ekUpdate the weights ω between network input layer and hidden layerij, hidden layer and
Weights ω between output layerjkFormula is as follows：
ωjk=ωjk+ηHjek
9.5) renewal of threshold value：Network node threshold value a, b are updated according to error e；
bk=bk+ek
9.6) weights, threshold value after adjusting assign improvement BP-MEA neutral nets, repeat step 9.1 again) -9.5) until
Error exceedes untill setting the upper limit less than training objective or cycle-index, and improvement BP-MEA neural metwork trainings are completed.
The present invention technical concept be：Mind evolutionary can accomplish collective search compared with traditional algorithm, evolve and calculate
Method has been applied to solve complicated forecast model, and plays good effect.Present invention improves over mind evolutionary, propose
Using the method for the BP artificial neural networks for improving mind evolutionary (MEA), two kinds are used simultaneously to the composition of colony
Method, a kind of is that random search forms new interim colony in global scope；It is another, it is to choose the higher individual of score
Between carry out cross and variation generation.Wherein, former is used in global scope search optimal solution, and the latter in local scope then to obtaining
To relative optimal solution intersected to seek obtaining more excellent solution, thus computational accuracy is higher than Traditional Thinking evolution algorithm.
Based on this, the present invention proposes pre- using the method for the BP artificial neural networks for improving mind evolutionary (MEA)
Survey the mass ratio of induced-to-inducing air (ε) and outlet back pressure (p of injectorc) etc. key parameter, can be without considering complicated flow mechanism, you can
Quickly and easily obtain it is high-precision predict the outcome, be injector it is related manufacture and design, circulating research etc. provides necessary base
Plinth.
Beneficial effects of the present invention are mainly manifested in：Working fluid pressure and driving fluid pressure only need to be measured using the present invention
Power, it is possible to be conveniently predicted to injector performance, effectively solves the problem of conventional method error is universal larger, carries
High precision of prediction.
Brief description of the drawings
Fig. 1 is BP neural network topological structure；
Fig. 2 is to improve BP-MEA neural network prediction resultant error figures；
Fig. 3 is to improve BP-MEA neural network prediction injector performance flow charts.
Embodiment
The invention will be further described below in conjunction with the accompanying drawings.
1~Fig. 3 of reference picture, a kind of injector based on using the BP artificial neural networks for improving mind evolutionary
The Forecasting Methodology of energy, comprises the following steps：
Step one, the collection and processing of data：For given injector, it is citation jet body pressure to collect relevant parameter
(Pe), working fluid pressure (Pp), outlet back pressure (Pc) and mass ratio of induced-to-inducing air (ε).To accelerate the convergence of neutral net and reducing instruction
Practice the time, it is necessary to citation jet body pressure, working fluid pressure and outlet back pressure are normalized, make its to [0,1] it
Between, formula is as follows：
Wherein, k is the data after normalization, and x is to be normalized data, xminTo be normalized the minimum value in data,
xmaxTo be normalized maximum in data.
Step 2, mind evolutionary initiation parameter is set：To including Population Size, winning sub- population number, interim
The parameters such as sub- population number, sub-population size, iterations are configured and set up initial population.
Step 3, several winning individuals and temporary individual of highest scoring in initial population are calculated according to scoring function,
New individual is produced centered on these individuals, winning sub- population and interim sub- population is obtained.
Individual in step 4, operation similartaxis, each sub- population is vied each other, until ripe sub- population is produced, by score
Recorded.
Step 5, operation dissimilation, the optimum individual of each ripe sub- population carries out global competition and forms new interim group
Body.
Step 6, the sub-group for keeping score of a relatively high in the relatively low sub-group of score progress cross and variation is obtained newly
Individual, then remaining sub-group is eliminated and re-started in global scope search for and form new interim colony.
Step 7, four~step 6 of repeat step reaches the iterations upper limit, and the scoring highest individual now produced is i.e.
For required optimized individual.
Step 8, neural network structure is set up and parameter initialization：Artificial neural network is obtained according to input and output vector
Topological structure, determine neural network structure input layer be 2, output layer neuron be 2, imply the number of plies be one
Layer, neuron number is 5, and required optimized individual is decoded, and extracts weights and threshold value carries out the foundation of artificial neural network.
Step 9, improves the training of BP-MEA neutral nets：Instructed with the artificial neural network set up in step (8)
Practice, by iteration several times, until predicted value and desired value are less than set-point, now neural metwork training terminates.The training is complete
Into neutral net be that the injector performance set up of the present invention predicts BP-MEA neutral nets.
Step 10, in practical implementation, the measured data of the given injector of collection, including citation jet body pressure
(Pe), working fluid pressure (Pp)；Data are normalized by the method for step one, foundation is then input to again and is completed
Improvement BP-MEA neutral nets in, obtain output vector outlet back pressure (Pc) and mass ratio of induced-to-inducing air (ε), then outlet back pressure is carried out
Renormalization processing, that is, obtain predicted value.
X=k (xmax-xmin)+xmin
In the step 2, by mind evolutionary parameter initialization, Population Size is 1000, winning sub- population number with
Interim sub- population number is 5, and every sub- population includes 100 individuals, and each individual is made up of the weights of neutral net with threshold value,
Because this BP-MEA neural network structures are that 2-5-2, i.e. input layer have 2 nodes, output layer has 2 nodes, and hidden layer has 5
Node, so 2 × 5 × 2=20 weights are had, 5+2 threshold value, so this algorithm individual UVR exposure length is 27 and weights and threshold
It is worth for the random number between [- 1,1].Evaluation function is set up using above-mentioned coding, individual is scored, wherein, score higher
Predicting the outcome for the BP-MEA neutral nets set up with the individual by weights and threshold value is more accurate.
In the step 3,1000 individuals are produced, is scored and is arranged according to score size ascending order according to step 3,
5 winning individuals of scoring highest and the higher temporary individual of 5 scorings are produced, centered on above-mentioned individual, in each individual
Surrounding produces 99 new individuals, so as to obtain the 5 winning sub- populations and 5 interim sub- populations that number of individuals is 100；
In the step 4, operation similartaxis process is：Individual in each sub- population is vied each other, and is chosen and is recorded optimal solution
(scoring highest) is produced new victor, if no longer producing new victor, and the sub-group is ripe sub-group, convergent process
Ripe winning population and ripe interim population are formed after end, the score that optimum individual in each sub- population is recorded respectively is used as this
The score of sub- population.
In the step 5, operation dissimilation process is：The optimum individual of each ripe sub- population carries out global competition, if facing
When sub-group score of the score higher than some ripe winning sub-group, then the interim subgroup that the winning sub-group can be won
Body is substituted, and the individual in former winning sub-group is released；If the score of a ripe interim sub-group is excellent less than any one
Win the score of sub-group, then the interim sub-group is gone out of use, and individual therein is released.
In the step 6, the interim sub-group eliminated is discharged and scans for re-forming newly in global scope
Interim sub-group, show optimum individual in two poor sub-groups in winning sub-group and intersected using real number interior extrapolation method
Operation, by k-th of chromosome a in the colonykWith l-th of chromosome alIt is as follows in the crossover operation methods of j：
akj=akj(1-r)+aljr
alj=alj(1-r)+akjr
Wherein, r is the random number between [0,1].Produce new individual and 99 new are produced using new individual as center
Body simultaneously carries out operation similartaxis and obtains carrying out operation dissimilation after ripe sub-group again.
In the step 7, the iterations upper limit 500 times, the scoring highest individual of generation is required optimized individual.
In the step 8：Improve the foundation of BP-MEA neural network structures and the initialization of relevant parameter：By optimized individual
Comprising threshold value decoded with weights, build BP artificial neural networks, according to neutral net input sample (driving fluid pressure
Power, working fluid pressure), output sample (outlet back pressure, mass ratio of induced-to-inducing air), between input layer, hidden layer and each layer of output layer
Connection weight initialization value takes [- 1,1] at random, uses ωij、ωjkRepresent, learning rate η takes 0.1, and training objective takes 0.00001, follows
Ring number of times 200 times.
In the step 9, input training sample starts to be trained process to neutral net as follows：
9.1) calculating of hidden layer：Wherein i, j, k represent input layer number, node in hidden layer and output layer section respectively
Points, f (x) is that transmission function takes S types (Sigmoid) function, and x is the data that output layer is inputted.
The input of hidden layer node
The output H of hidden layer nodej=f (Sj)
9.2) calculating of output layer：Wherein YbFor the prediction output during neural network.
Export the output of node layer
9.3) error calculation：K-th of the neuron prediction of neutral net output layer is output as Ybk, YkFor k-th neuron
, there is error e between them in desired outputk, formula is as follows：
ek=Yk-Ybk
9.4) renewal of weights：According to error ekUpdate the weights ω between network input layer and hidden layerij, hidden layer and
Weights ω between output layerjkFormula is as follows：
ωjk=ωjk+ηHjek
9.5) renewal of threshold value：Network node threshold value a, b are updated according to error e.
bk=bk+ek
9.6) weights, threshold value after adjusting assign improvement BP-MEA neutral nets, repeat step 9.1 again) -9.5) until
Error exceedes untill setting the upper limit less than training objective or cycle-index, and improvement BP-MEA neural metwork trainings are completed.
Example：Preferably to embody the effect of the present invention, the method for the present invention is now subjected to actual motion.Using document 3
(IW.Eames et al.A theoretical and experimental study of a small-scale steam
jet refrigerator.International journal of refrigeration,18(6):378-386,1995, i.e.,
The theoretical and experimental study International refrigeration journals of the small steam ejector refrigeration machines such as IW.Eames, 18 (6):378-386,
1995) method in obtains citation jet body pressure in 706Pa-2339Pa 110 groups of data, and 80 groups of data are randomly selected therefrom
As training sample, carry out repeatedly training with this patent methods described and obtain BP neural network.Remaining 30 groups of data are used for
The reliability of neutral net is verified, input sample (citation jet body pressure p is chosene, working fluid pressure pp) finished using training
BP-MEA neutral nets carry out outlet back pressure pcWith mass ratio of induced-to-inducing air ε prediction, the result of prediction and result by references are compared,
Calculation error, formula is as follows：
Wherein, μ is predicted value and the error of literature value, and neural network prediction value is Yb, Y is literature value.
Specifically predict the outcome with literature value as shown in table 1 and Fig. 2,
Table 1
Wherein, the mean error finally predicted the outcome is 0.28%, and worst error is 1.1%, refreshing using BP-MEA is improved
Predicting the outcome through network is more accurate.Improve the pre- of injector mass ratio of induced-to-inducing air that BP-MEA neutral nets complete and outlet back pressure
Method in survey method and conventional method such as document 1 and document 2 is compared it can be found that using the average mistake of conventional model
It is poor many in 5-10%, and worst error is up to more than 15%.It can be seen that the method for use this patent is ensureing what is be quickly predicted
Under the premise of, precision of prediction can be greatly promoted.
Claims (9)
1. a kind of Forecasting Methodology of the injector performance based on using the BP artificial neural networks for improving mind evolutionary, its
It is characterised by：Comprise the following steps：
Step one, the collection and processing of data：For given injector, it is citation jet body pressure Pe, work to collect relevant parameter
Make Fluid pressure Pp, outlet back pressure Pc and mass ratio of induced-to-inducing air ε, citation jet body pressure, working fluid pressure and outlet back pressure are carried out
Normalized, makes it arrive between [0,1], formula is as follows：
Wherein, k is the data after normalization, and x is to be normalized data, xminTo be normalized the minimum value in data, xmaxFor
It is normalized maximum in data；
Step 2, mind evolutionary initiation parameter is set：To being planted including Population Size, winning sub- population number, interim son
The parameters such as group's number, sub-population size, iterations are configured and set up initial population；
Step 3, several winning individuals and temporary individual of highest scoring in initial population is calculated according to scoring function, with this
New individual is produced centered on a few bodies, winning sub- population and interim sub- population is obtained；
Individual in step 4, operation similartaxis, each sub- population is vied each other, until producing ripe sub- population, score is carried out
Record；
Step 5, operation dissimilation, the optimum individual of each ripe sub- population carries out global competition and forms new interim colony；
Step 6, carries out cross and variation by the sub-group for keeping score of a relatively high in the relatively low sub-group of score and obtains new
Body, then remaining sub-group is eliminated and search is re-started in global scope and new interim colony is formed；
Step 7, four~step 6 of repeat step reaches the iterations upper limit, and the scoring highest individual now produced is institute
Seek optimized individual；
Step 8, neural network structure is set up and parameter initialization：According to the input and output vector of network, neutral net knot is determined
Structure, it is one layer to imply the number of plies, neuron number rule of thumb formula：Draw, wherein l, n are input, exported
Nodes, a is constant；Wherein, citation jet body pressure, working fluid pressure are the input of the neutral net, outlet back pressure, injection
Coefficient is the output of the neutral net；
Step 9, improves the training of BP-MEA neutral nets：It is trained with the artificial neural network set up in step (8),
By iteration several times until predicted value and desired value are less than set-point, now neural metwork training terminates, what the training was completed
Neutral net is the injector performance prediction BP-MEA neutral nets of foundation；
Step 10, in practical implementation, the measured data of the given injector of collection, including citation jet body pressure Pe, work
Fluid pressure Pp；Data are normalized by the method for step one, the improvement BP- for setting up and completing then is input to again
In MEA neutral nets, output vector outlet back pressure Pc and mass ratio of induced-to-inducing air ε are obtained, then outlet back pressure is subjected to renormalization processing,
Obtain predicted value.
2. the injector performance as claimed in claim 1 based on using the BP artificial neural networks for improving mind evolutionary
Forecasting Methodology, it is characterised in that：By mind evolutionary parameter initialization in the step 2, such as Population Size, winning son
Population number and interim sub- population number, each individual are then made up of the weights and threshold value of neutral net；BP-MEA nerve nets
Network structure is that l-m-n, i.e. input layer have l node, and hidden layer has m node, and output layer has n node, so shared l × m
× n weights, m+n threshold value, individual UVR exposure length is l × m × n+m+n, and weights are random between [- 1,1] with threshold value
Number；Evaluation function is set up using above-mentioned coding, individual is scored, scoring is higher to be set up with the individual by weights and threshold value
BP-MEA neutral nets predict the outcome it is more accurate.
3. the injector as claimed in claim 1 or 2 based on using the BP artificial neural networks for improving mind evolutionary
The Forecasting Methodology of performance, it is characterised in that：In the step 3, produce S individual, scored according to step 3 and according to must
Divide the arrangement of size ascending order, produce N1The individual winning individual of scoring highest and N2The higher temporary individual of individual scoring, using above-mentioned individual as
Center, produces H new individuals around each individual, so as to obtain the N that number of individuals is H1Individual winning sub- population and N2It is individual interim
Sub- population.
4. the injector as claimed in claim 1 or 2 based on using the BP artificial neural networks for improving mind evolutionary
The Forecasting Methodology of performance, it is characterised in that：In the step 4, operation similartaxis process is：Individual in each sub- population is mutually competing
Strive, choose and record optimal solution for produced new victor, if no longer producing new victor, the sub-group is ripe subgroup
Body, convergent process forms ripe winning population and ripe interim population after terminating, optimum individual in each sub- population is recorded respectively
Score as the sub- population score.
5. the injector as claimed in claim 1 or 2 based on using the BP artificial neural networks for improving mind evolutionary
The Forecasting Methodology of performance, it is characterised in that：In the step 5, operation dissimilation process is：Optimal of each ripe sub- population
Body carries out global competition, if score of the score higher than some ripe winning sub-group of interim sub-group, the winning subgroup
Know from experience the interim sub-group won to substitute, the individual in former winning sub-group is released；If a ripe interim sub-group
Score be less than any one winning sub-group score, then the interim sub-group go out of use, individual therein is released.
6. the injector as claimed in claim 1 or 2 based on using the BP artificial neural networks for improving mind evolutionary
The Forecasting Methodology of performance, it is characterised in that：In the step 6, the interim sub-group eliminated is discharged and in global scope
Scan for re-forming and optimum individual use in preferable two sub-groups is showed in new interim sub-group, winning sub-group
Real number interior extrapolation method carries out crossover operation, by k-th of chromosome a in the colonykWith l-th of chromosome alThe crossover operation side of j
Method is as follows：
akj=akj(1-r)+aljr
alj=alj(1-r)+akjr
Wherein, r is the random number between [0,1], produces new individual and S new individuals are produced using new individual as center simultaneously
Operation similartaxis is carried out to obtain carrying out operation dissimilation after ripe sub-group again.
7. the injector as claimed in claim 1 or 2 based on using the BP artificial neural networks for improving mind evolutionary
The Forecasting Methodology of performance, it is characterised in that：In the step 7, the iterations upper limit, the scoring highest individual of generation are set
As required optimized individual.
8. the injector as claimed in claim 1 or 2 based on using the BP artificial neural networks for improving mind evolutionary
The Forecasting Methodology of performance, it is characterised in that：In the step 8, improve BP-MEA neural network structures and set up and relevant parameter
Initialization：The threshold value that optimized individual is included is decoded with weights, builds BP artificial neural networks, defeated according to neutral net
Enter sample, i.e. citation jet body pressure, working fluid pressure, export sample, i.e. outlet back pressure, mass ratio of induced-to-inducing air, input layer, hidden layer
Connection weight initialization value between each layer of output layer takes [- 1,1] at random, uses ωij、ωjkRepresenting, learning rate η takes 0.1~
0.2, training objective takes 10-3 ~10-6。
9. the injector as claimed in claim 1 or 2 based on using the BP artificial neural networks for improving mind evolutionary
The Forecasting Methodology of performance, it is characterised in that：In the step 9, input training sample starts to be trained process to neutral net
It is as follows：
9.1) calculating of hidden layer：Wherein i, j, k represent input layer, hidden layer node and output node layer, f (x) respectively
S type functions are taken for transmission function, x is the data that output layer is inputted；
The input of hidden layer node
The output H of hidden layer nodej=f (Sj)
9.2) calculating of output layer：Wherein YbFor the prediction output during neural network；
Export the output of node layer
9.3) error calculation：K-th of the neuron prediction of neutral net output layer is output as Ybk, YkFor the expectation of k-th of neuron
, there is error e between them in outputk, formula is as follows：
ek=Yk-Ybk
9.4) renewal of weights：According to error ekUpdate the weights ω between network input layer and hidden layerij, hidden layer and output
Weights ω between layerjkFormula is as follows：
ωjk=ωjk+ηHjek
9.5) renewal of threshold value：Network node threshold value a, b are updated according to error e；
bk=bk+ek
9.6) weights, threshold value after adjusting assign improvement BP-MEA neutral nets, repeat step 9.1 again) -9.5) until error
Exceed less than training objective or cycle-index untill the upper limit is set, improvement BP-MEA neural metwork trainings are completed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710111923.8A CN106991208A (en) | 2017-02-28 | 2017-02-28 | Forecasting Methodology based on the injector performance using the BP artificial neural networks for improving mind evolutionary |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710111923.8A CN106991208A (en) | 2017-02-28 | 2017-02-28 | Forecasting Methodology based on the injector performance using the BP artificial neural networks for improving mind evolutionary |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106991208A true CN106991208A (en) | 2017-07-28 |
Family
ID=59412596
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710111923.8A Pending CN106991208A (en) | 2017-02-28 | 2017-02-28 | Forecasting Methodology based on the injector performance using the BP artificial neural networks for improving mind evolutionary |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106991208A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107610009A (en) * | 2017-08-21 | 2018-01-19 | 浙江工业大学 | A kind of Trinity College Enrollment probability forecasting method based on neutral net |
| CN108197748A (en) * | 2018-01-08 | 2018-06-22 | 国网江苏省电力有限公司 | A kind of flow of the people Forecasting Methodology based on mind evolutionary |
| CN114047413A (en) * | 2021-11-11 | 2022-02-15 | 国网上海市电力公司 | GIS partial discharge identification method and system based on MIV and MEA-LVQ neural network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040167721A1 (en) * | 2001-07-27 | 2004-08-26 | Masahiro Murakawa | Optimal fitting parameter determining method and device, and optimal fitting parameter determining program |
| CN102222267A (en) * | 2010-04-14 | 2011-10-19 | 同济大学 | Remote-sensing classification processing method for neural network by genetic algorithm |
| CN105046355A (en) * | 2015-07-13 | 2015-11-11 | 国家电网公司 | System energy consumption modeling method based on improved mind evolutionary algorithm |
| CN105913150A (en) * | 2016-04-12 | 2016-08-31 | 河海大学常州校区 | BP neural network photovoltaic power station generating capacity prediction method based on genetic algorithm |
-
2017
- 2017-02-28 CN CN201710111923.8A patent/CN106991208A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040167721A1 (en) * | 2001-07-27 | 2004-08-26 | Masahiro Murakawa | Optimal fitting parameter determining method and device, and optimal fitting parameter determining program |
| CN102222267A (en) * | 2010-04-14 | 2011-10-19 | 同济大学 | Remote-sensing classification processing method for neural network by genetic algorithm |
| CN105046355A (en) * | 2015-07-13 | 2015-11-11 | 国家电网公司 | System energy consumption modeling method based on improved mind evolutionary algorithm |
| CN105913150A (en) * | 2016-04-12 | 2016-08-31 | 河海大学常州校区 | BP neural network photovoltaic power station generating capacity prediction method based on genetic algorithm |
Non-Patent Citations (3)
| Title |
|---|
| 孙俊 等: "基于MEA-BP神经网络的大米水分含量高光谱技术检测", 《食品科学》 * |
| 李晓敏: "一种包含遗传算子的思维进化算法", 《数字技术与应用》 * |
| 黄亮亮: "基于人工神经网络的喷射器性能预测及优化研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107610009A (en) * | 2017-08-21 | 2018-01-19 | 浙江工业大学 | A kind of Trinity College Enrollment probability forecasting method based on neutral net |
| CN107610009B (en) * | 2017-08-21 | 2021-06-18 | 浙江工业大学 | Trinity enrollment probability prediction method based on neural network |
| CN108197748A (en) * | 2018-01-08 | 2018-06-22 | 国网江苏省电力有限公司 | A kind of flow of the people Forecasting Methodology based on mind evolutionary |
| CN108197748B (en) * | 2018-01-08 | 2020-11-24 | 国网江苏省电力有限公司 | People flow prediction method based on thought evolution algorithm |
| CN114047413A (en) * | 2021-11-11 | 2022-02-15 | 国网上海市电力公司 | GIS partial discharge identification method and system based on MIV and MEA-LVQ neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106951983A (en) | Injector performance Forecasting Methodology based on the artificial neural network using many parent genetic algorithms | |
| CN106980897A (en) | A kind of injector performance parameter prediction method of the BP artificial neural networks based on learning rate changing | |
| CN106960217A (en) | The Forecasting Methodology of injector performance based on the BP artificial neural networks using depth Adaboost algorithm | |
| CN104361178A (en) | Indoor evacuation simulating optimization method based on potential energy driving cellular ant colony algorithm | |
| CN106991208A (en) | Forecasting Methodology based on the injector performance using the BP artificial neural networks for improving mind evolutionary | |
| CN108036790B (en) | Robot path planning method and system based on ant-bee algorithm in obstacle environment | |
| CN109376801B (en) | Wind driven generator blade icing diagnosis method based on integrated deep neural network | |
| CN107092255A (en) | A kind of multi-robots path-planning method based on improved adaptive GA-IAGA | |
| CN106960075A (en) | The Forecasting Methodology of the injector performance of RBF artificial neural network based on linear direct-connected method | |
| CN109165664A (en) | A kind of attribute missing data collection completion and prediction technique based on generation confrontation network | |
| CN110851566B (en) | Differentiable network structure searching method applied to named entity recognition | |
| CN106067066A (en) | Method for diagnosing fault of power transformer based on genetic algorithm optimization pack algorithm | |
| CN109634121A (en) | More parent genetic algorithm air source heat pump multiobjective optimization control methods based on radial basis function neural network | |
| CN114692265B (en) | Zero-carbon building optimization design method based on deep reinforcement learning | |
| CN107414830A (en) | A kind of carrying machine human arm manipulation multi-level mapping intelligent control method and system | |
| CN106597154B (en) | Transformer fault diagnosis method for improving based on DAG-SVM | |
| CN110765582B (en) | Self-organization center K-means microgrid scene division method based on Markov chain | |
| CN104680257B (en) | Towards the intelligent optimization method of the oil pumper course of work | |
| CN111625992A (en) | Mechanical fault prediction method based on self-optimization deep learning | |
| CN105389614B (en) | A kind of implementation method of neutral net self refresh process | |
| CN104680259B (en) | The energy conservation optimizing method of oil extractor operating parameter | |
| Londoño et al. | Graph Grammars for Super Mario Bros Levels. | |
| CN106503793B (en) | A kind of neural network short-term wind speed forecasting method based on improvement difference algorithm | |
| Manivannan et al. | Convolutional neural network and feature encoding for predicting the outcome of cricket matches | |
| CN102324059A (en) | Target assignment method based on evolution |
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 |
Application publication date: 20170728 |
|
| RJ01 | Rejection of invention patent application after publication |