CN109409568A - A kind of prediction technique based on genetic algorithm optimization BP neural network underground water buried depth - Google Patents
A kind of prediction technique based on genetic algorithm optimization BP neural network underground water buried depth Download PDFInfo
- Publication number
- CN109409568A CN109409568A CN201811092233.3A CN201811092233A CN109409568A CN 109409568 A CN109409568 A CN 109409568A CN 201811092233 A CN201811092233 A CN 201811092233A CN 109409568 A CN109409568 A CN 109409568A
- Authority
- CN
- China
- Prior art keywords
- neural network
- individual
- genetic algorithm
- underground water
- buried depth
- 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
- 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"
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Mathematical Physics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Economics (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Genetics & Genomics (AREA)
- Physiology (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides a kind of prediction technique based on genetic algorithm optimization BP neural network underground water buried depth, and the method includes at least: the acquisition of data, and Set Pair Analysis screens independent variable, takes the corresponding sample of the target independent variable；The setting of BP neural network obtains the output result of BP neural network；Genetic algorithm initial parameter is set；Individual choice operation: it is used to generate the next generation, selection principle with predetermined probabilities selection individual in old group are as follows: the fitness value according to individual is selected from big to small；Crossover operation: by the combined crosswise of chromosome, new individual is generated；Mutation operation: referring to an optional individual from group, makes a variation certain section of coding in selected chromosome to generate new individual；Reach evolution maximum number of times, calculates fitness, group is compared with original seed；The corresponding chromosome of fitness optimal solution is threshold value corresponding to BP neural network and weight.Using the embodiment of the present invention, the accuracy of underground water buried depth can be improved.
Description
Technical field
The present invention relates to a kind of underground water buried depth electric powder predictions, are based on genetic algorithm optimization BP more particularly to one kind
The prediction technique of neural network underground water buried depth.
Background technique
Ground water regime is one and is related to multifactor complication system, by atmospheric circulation, solar activity, hydrometeorology early period
The combined influence of numerous physical agents such as element, region cover variation, the short-term forecast of underground water buried depth is regional water safety pipe
The important topic that the important content and hydroscience circle of reason study for a long period of time.In recent years, BP neural network and Set Pair Analysis are provided in water
It is widely used and promotes in terms of source, Wu Hongbin establishes multiple linear regression model to Zunyi sea otter dam underground water
The dynamic prediction of water quality, the results showed that precision of prediction is higher, and the model of foundation is more conform with the status in research area.Dong plays wide etc. build
Vertical BP neural network model carries out buried depth prediction to Weihe north two inspection well points with typicalness of non-irrigated source region, as the result is shown BP mind
Through network compared with linear regression model (LRM) to the dynamic model of nonlinear change have preferable learning ability, precision of prediction compared with
It is high.Chi Baoming etc. establishes GA-BP neural network prediction model and carries out dynamic to the underground water of 6 monitoring wells of ingot Open pit Area
Prediction, the results showed that GA-BP Neural Network model predictive precision is substantially better than BP neural network model.As can be seen that with mould
The continuous improvement of type, precision of prediction also greatly improve therewith.On the other hand, the screening of independent variable has prediction in the prediction of model
Vital effect.Tong Li etc. has studied the select permeability of large-scale parameter model, for the situation of model morbid state, proposes
The Cp criterion of independent variable screening based on Principal Component Estimation, is remarkably improved in systematic error and measurement error.King
Silver-colored brightness etc. considers the influence that Variable selection generates predicted value, discusses dependent variable with risk function in linear model
The Optimality of prediction.Liu Liyang excludes some influences not by there is the variable of remarkable effect that regression equation is added to dependent variable
Significant variable, establishes one " optimal " independent variable subset, and the studies above demonstrates screening independent variable to precision of prediction
Validity.With the understanding that deepens constantly of prediction model continuously improved with forecasting mechanism, the precision and reliability of prediction are obtained
Very big promotion, the above research have established important basis to underground water buried depth prediction model and mechanism identification.
With the understanding that deepens constantly of prediction model continuously improved with forecasting mechanism, the precision and reliability of prediction are obtained
Very big promotion.But the research of current underground water buried depth prediction is focused on new model method mostly and is introduced, and it is less from underground
The screening of water independent variable improves input factor representativeness and sets out, and the screening of prediction model independent variable is to improve the prediction of underground water buried depth
Horizontal important foundation may be affected to the result of prediction if the representativeness that the sample of input layer is chosen is lower, how will
It is crucial that representative biggish independent variable, which remains,.There is to initial weight very sensitive, pole in neural network common simultaneously
Local minimum easily is converged on, the deficiencies of ability of searching optimum is poor.Although BP neural network algorithm can after adaptive impovement
The problems such as being easily trapped into slow convergence rate, local minimum of BP network is solved, but cannot still overcome BP completely in practical applications
The intrinsic defect of neural network algorithm.
Summary of the invention
In view of the foregoing deficiencies of prior art, the present invention chooses from underground water buried depth prediction model in independent variable subjective
Property relatively strong, prediction model the problems such as being easily trapped into local optimum angle set out, propose based on Set Pair Analysis and genetic optimization BP
Neural network prediction model.Set Pair Analysis describes the connection between underground water buried depth and all kinds of influence factors in terms of same, different, anti-three
Degree of being, preferably the higher independent variable of Pair Analysis is used as input out；It is pre- to underground water using genetic algorithm optimization BP neural network again
Model is surveyed to optimize.The embodiment of the present invention is based on the BP neural network after Set Pair Analysis and genetic algorithm optimization compared with BP nerve net
Network and the genetic algorithm optimization BP neural network superior performance model for not doing independent variable screening, can quick and precisely carry out underground water
The prediction of buried depth.
In order to achieve the above objects and other related objects, the present invention provides a kind of based on genetic algorithm optimization BP neural network
The prediction technique of underground water buried depth, the method include at least:
(1) acquisition of data
It is selected with the higher independent variable of areal association degree to be measured of underground water buried depth using Method of Set Pair Analysis as target
Independent variable, and obtain the corresponding sample of the target independent variable；
(11) setting of BP neural network
Using the corresponding sample of the target independent variable as the training sample of BP neural network, the defeated of BP neural network is determined
The quantity of ingress, output node and hidden layer obtains the output result of BP neural network；
(2) genetic Optimization Algorithm
(21) genetic algorithm initial parameter is set
Population scale, evolution number, crossover probability and mutation probability are set, the weight of BP neural network and threshold value are formed
Genetic coding；
(22) building of fitness function
Using the function that BP neural network prediction output is formed with desired output as fitness value；
(23) genetic algorithm treatment process
(231) individual choice operates: being used to generate the next generation, selection principle with predetermined probabilities selection individual in old group
Are as follows: the fitness value according to individual is selected from big to small；
(232) crossover operation: refer to the foundation for selecting a plurality of parent chromosome to carry out next-generation chromosome from population, lead to
The combined crosswise for crossing chromosome generates new individual；
(233) mutation operation: referring to an optional individual from group, carries out to certain section coding in selected chromosome
Variation is to generate new individual；
(234) step (231) to step (233) are repeated, constantly individual in population is selected, is intersected, make a variation behaviour
Make and record fitness value, reach evolution maximum number of times, the genetic coding in the individual newly obtained is decoded, calculates and adapt to
Degree, group is compared with original seed；The corresponding chromosome of fitness optimal solution is weight corresponding to BP neural network and threshold value；
(3) training process of BP neural network
The weight and threshold value obtained using step (234) is trained BP neural network, until determining obtained error
Less than there is preset error value.
In the preferred embodiment of the present invention, the BP neural network termination condition setting are as follows:
Frequency of training 5000, convergence error 0.001, input layer have 10 nodes, and hidden layer has 19 nodes, and output layer has
1 node.
In the preferred embodiment of the present invention, the genetic algorithm parameter setting are as follows:
Evolution number is 40 times；Population scale n value range is 20~80, and taking population scale is 30, and crossover probability is
0.2；Mutation probability is 0.1.
It is described higher with areal association degree to be measured using Method of Set Pair Analysis selection in the preferred embodiment of the present invention
The specific formula that independent variable is taken as the step of target independent variable are as follows:
Setting first set A indicates that independent variable, second set B indicate the measured value in area to be measured, and first set A and the
Two set B composition integrates to as H:
H=(A, B)
Wherein, A and B has N number of characteristic features, comprising: S denominator, P opposed characteristics, F difference characteristic.
If enabling a=S/N, b=F/N, c=P/N, then collecting indicates the Pair Analysis (frequently referred to three-unit connection number) of H are as follows:
U=a+bI+cJ
I=a/ (a+c)+c/ (a+c) J
Wherein, U is Pair Analysis；A is unification degree；B is diversity factor；C is opposition degree；And a+b+c=1； I
For difference label symbol, I ∈ [- 1,1]；J is opposition label symbol, and value is -1.
In the preferred embodiment of the present invention, the code length of the genetic algorithm is
S=n × m+m × l+l
In formula, m is node in hidden layer；N is input layer number；L output layer number of nodes.
In the preferred embodiment of the present invention, the fitness function is embodied are as follows:
In formula, SE is the error sum of squares between the prediction output of neural network and desired output.
In the preferred embodiment of the present invention, individual select probability value is embodied are as follows:
In formula, PiFor the select probability value of individual i, fiFor the fitness value of individual i；N is population at individual number.
As described above, the prediction technique of the invention based on genetic algorithm optimization BP neural network underground water buried depth, has
Below the utility model has the advantages that
(1) present invention, can using the independent variable preferably gone out as input on the basis of Set Pair Analysis optimal independent variables
The correlation of independent variable and the prediction of underground water buried depth is improved, and then effectively improves precision of prediction；Construct ground using BP neural network
The prediction model being lauched, and it is optimized, establish the ground based on Set Pair Analysis and genetic algorithm optimization BP neural network
It is lauched the prediction model of buried depth, determines that the optimal weight and threshold value in BP neural network gives BP nerve net by genetic algorithm
Network, and underground water buried depth is predicted using the BP neural network, the accuracy of prediction is improved, prediction error is reduced.
(2) on the basis of analysis and understanding BP neural network algorithm and genetic algorithm idea, establishing structure is 10-19-1
Neural network, by verification experimental verification as can be seen that based on Set Pair Analysis and BP neural network training stage it is relatively average accidentally
Difference is 0.1210, forecast period relative average error 0.1508；The genetic algorithm optimization BP neural network of independent variable screening is not done
The relative average error of training stage is 0.0867, the relative average error 0.0943 of forecast period；Based on Set Pair Analysis and something lost
The relative average error of propagation algorithm Optimized BP Neural Network training stage is 0.0798, forecast period relative average error
0.0884.Three kinds of model results are shown: precision of the genetic algorithm optimization BP neural network to training and the prediction of underground water buried depth
Preferably, but it is higher based on Set Pair Analysis and genetic algorithm optimization BP neural network precision of prediction, meet actual requirement, is underground water
The prediction of buried depth provides effective method.
Detailed description of the invention
Fig. 1 is a kind of case study on implementation of connection angle value of the embodiment of the present invention；
Fig. 2 is BP neural network hidden layer when being 15-30 relative average error figure when predicting；
Fig. 3 is based on Set Pair Analysis and BP neural network prediction result；
Fig. 4 is entire variable genetic algorithm optimization BP neural network prediction result；
Fig. 5 is based on Set Pair Analysis and genetic algorithm optimization BP neural network algorithm flow chart；
Fig. 6 is based on Set Pair Analysis and genetic algorithm optimization BP neural network prediction result；
Fig. 7 is based on Set Pair Analysis and genetic algorithm optimization BP neural network prediction error.
Specific embodiment
Illustrate embodiments of the present invention below by way of specific specific example, those skilled in the art can be by this specification
Other advantages and efficacy of the present invention can be easily understood for disclosed content.The present invention can also pass through in addition different specific realities
The mode of applying is embodied or practiced, the various details in this specification can also based on different viewpoints and application, without departing from
Various modifications or alterations are carried out under spirit of the invention.
Please refer to Fig. 2-7.It should be noted that only the invention is illustrated in a schematic way for diagram provided in the present embodiment
Basic conception, with only show in schema with related component in the present invention rather than component count, shape when according to actual implementation
Shape and size are drawn, when actual implementation kenel, quantity and the ratio of each component can arbitrarily change for one kind, and its component cloth
Office's kenel may also be increasingly complex.
(1) acquisition of data
Set Pair Analysis reflects the similitude between underground water buried depth historical sample in terms of same, different, anti-three, preferably goes out to contact
Higher independent variable is spent as input, and the independent variable preferably come out is pre-processed.
Using Method of Set Pair Analysis select with the regional higher independent variable of underground water buried depth Pair Analysis to be measured as target oneself
Variable, and obtain the corresponding sample of the target independent variable；The measured value of given two set independent variable A and area B to be predicted,
And collection that the two groups are combined into is set to H=(A, B), wherein A and B there are N number of characteristic features, comprising: S denominator, P
A opposed characteristic, F difference characteristic.If enabling a=S/N, b=F/N, c=P/N, then collect Pair Analysis (the frequently referred to ternary to H
Contact number) it indicates are as follows:
U=a+bI+cJ
In formula, U is Pair Analysis；A is unification degree；B is diversity factor；C is opposition degree；And a+b+c=1；I
For difference label symbol, I ∈ [- 1,1]；J is opposition label symbol, and J=-1 calculates I using ratio value method:
I=a/ (a+c)+c/ (a+c) J
It is opposite with training according to relative prediction residual criterion according to the actual conditions of local ground watering buried depth forecasting problem
Accidentally absolute value of the difference come description collections A and set B close to attribute, carry out same, different, anti-quantitative comparison and analyze, relative error
Absolute value using following formula calculate:
In formula, i=1~m, j=1~n.
The absolute value d of relative error is calculated with each single independent variable and the measured value in area to be predicted respectivelyI, j, invite special
Family determines the allowable error a of low requirement according to the calculated result of absolute relative error1With the allowable error a of high request2.In reality
In the calculated result of border area underground water buried depth background problems, relevant classification is as follows:
In formula, k is counter.
(1) setting of BP neural network
Using the corresponding sample of the target independent variable as the training sample of BP neural network, the defeated of BP neural network is determined
The quantity of ingress, output node and hidden layer obtains the output result of BP neural network.
The learning process of BP neural network model is made of forward and reverse propagation, the input letter of forward-propagating input sample
Breath, backpropagation transmission error and adjustment information.The training process of the model is divided into two parts: first is that forward-propagating process,
The input factor, which is successively calculated from input layer through hidden layer, is transmitted to output layer, exports corresponding input pattern in each neuron of output layer
Network response；Second is that back-propagation process, when output layer cannot get the desired output factor, error is transferred to backpropagation at this time,
By the error principle for reducing desired output and reality output, intermediate each layer is returned from output layer, input layer is eventually passed back to, repairs layer by layer
Just each connection weight and threshold value.With the lasting progress of error back propagation training, network responds input pattern correct
Rate is also continuously improved, and so circulation is gone down, either frequency of training reaches within the scope of the error of output reaches permission
Until the number being pre-designed.
1.3 foundation based on Set Pair Analysis and BP neural network underground water buried depth prediction model
Mengcheng County underground water buried depth data source is in Huaihe River Water Resources Commission Water Resources Research Institute of Anhui Water Resources, originally
Invention considers time and the two aspect factor of space, and 10 period underground water buried depths before Mengcheng County and previous hourly precipitation amount are made
For time factor, by the Fengtai County adjacent with Mengcheng County, Suzhou City, Huaiyuan County, Suixi County, Woyang County and Lixin County totally 6 ground
Area early period underground water buried depth as steric factor, total 17 factors are included in Set Pair Analysis independent variable optimization model together.It invites
Please expert the allowable error a of low requirement determined according to the calculated result that underground water buried depth absolute relative error is surveyed in Mengcheng County1
=0.1 and high request allowable error a2=0.2, the angle value that contacts between underground water buried depth measured value and independent variable reaches 0.8
When, it is believed that the independent variable representativeness is stronger.17 alternative independents variable are screened using Set-pair Analysis Model, for verifying collection pair
Analysis model, while calculating the related coefficient of Mengcheng County underground water buried depth and each independent variable, such as following formula:
In formula, A is regional measured value to be predicted；B is independent variable；N is the number of measured value.
Two groups of calculated results one are listed in Fig. 1, and calculate Mengcheng County underground water buried depth measured value and 17 alternative changes certainly
The related coefficient of amount can be concluded that
(1) Set-pair Analysis Model screens 17 alternative independents variable, and connection angle value is more than 0.8 to share 10, distinguishes
For Mengcheng County rainfall (t-1), first 1-5 months underground water buried depths, Lixin County underground water early period buried depth (t-1), Suixi
County's underground water early period buried depth (t-1), Woyang County underground water early period buried depth (t-1) and Huaiyuan County underground water early period buried depth (t-1), sieve
The related coefficient for 10 independents variable selected is only low with neighbouring area Woyang County underground water early period buried depth (t-1) related coefficient
In 0.8, other independent variable related coefficients have been more than 0.8.It can be seen that Method of Set Pair Analysis screening independent variable is believable.
(2) in terms of independent variable screening, Method of Set Pair Analysis is advantageous compared with correlation coefficient process: because Set-pair Analysis Model exists
When screening independent variable, the Pair Analysis between underground water buried depth measured value and independent variable can be calculated, while can also be calculated certainly
Pair Analysis between variable and underground water buried depth measured value, when the two connection angle value reaches certain standard, then it is assumed that should
Independent variable is stronger to this area's underground water buried depth representativeness.And the related coefficient of two variables is unrelated with variable sequence, in evaluation
It is relatively lower than Method of Set Pair Analysis in terms of physical meaning.
Proposed adoption list hidden layer structure of the present invention determines best node in hidden layer using the method for exhaustion.Rule of thumb, initially
The implicit number of plies is 15, and terminating the implicit number of plies is the BP neural network prediction-error image that 30, Fig. 2 is hidden layer node 15-30, can
When finding out node in hidden layer difference, prediction error also can be just different.Number of nodes is very little, and network cannot learn well,
Need to increase frequency of training, trained precision is also impacted；Number of nodes is too many, and the training time increases, and network is easy over-fitting.Cause
This tests the node in hidden layer of BP neural network, by comparing find node in hidden layer be 19 when prediction error compared with
It is small, therefore determine that node in hidden layer is 19, therefore BP network topology structure of the present invention is 10-19-1, therefore input layer has 10
Node, hidden layer have 19 nodes, and output layer has 1 node.
343 are selected in chronological order in 439 groups of inputoutput datas as training data and is used for network training, 96 groups of numbers
According to as test, it is normalized for the estimated performance of test b P neural network, and to inputoutput data.With collection pair
10 independents variable that analysis preferably goes out BP neural network prediction as input, Fig. 3 is with the BP neural network based on Set Pair Analysis
Prediction result.
It can be seen that using preferred 10 independents variable of Set Pair Analysis as input, when the implicit number of plies is 19, BP nerve net
The prediction result of network is best, and the relative average error of prediction is 0.1508, and prediction error is smaller, the predictability of neural network model
It can be preferably.
1.4 are based on Set Pair Analysis and genetic optimization BP neural network prediction model
Since BP neural network algorithm has the disadvantages of being easily trapped into local minimum point, with genetic algorithm to BP nerve
The characteristics of network optimizes, and basic thought is search of overall importance using genetic algorithm changes BP algorithm and relies on gradient information
The method for instructing to adjust neural network weight is found and is the most suitably connected to the network weight and network structure.The algorithm has
The global search technology of adaptive probability, breaks through the searching method of traditional rule, keeps the process of search more flexible, while in multimodal
The problem of on there is global assurance ability.
(2) genetic Optimization Algorithm
(21) genetic algorithm initial parameter is set
Population scale, evolution number, crossover probability and mutation probability are set, the weight of BP neural network and threshold value are formed
Genetic coding；
(22) building of fitness function
Using the function that BP neural network prediction output is formed with desired output as fitness value；
(23) genetic algorithm treatment process
(231) individual choice operates: being used to generate the next generation, selection principle with predetermined probabilities selection individual in old group
Are as follows: the fitness value according to individual is selected from big to small；
(232) crossover operation: refer to the foundation for selecting a plurality of parent chromosome to carry out next-generation chromosome from population, lead to
The combined crosswise for crossing chromosome generates new individual；
(233) mutation operation: referring to an optional individual from group, carries out to certain section coding in selected chromosome
Variation is to generate new individual；
(234) step (231) to step (233) are repeated, constantly individual in population is selected, is intersected, make a variation behaviour
Make and record fitness value, reach evolution maximum number of times, the genetic coding in the individual newly obtained is decoded, calculates and adapt to
Degree, group is compared with original seed；The corresponding chromosome of fitness optimal solution is threshold value corresponding to BP neural network and weight；
The present invention to individual use the mode of real coding to encode: code length for
S=n × m+m × l+l
In formula, m is node in hidden layer；N is input layer number；L output layer number of nodes.
Population scale has a great impact to the global search performance of genetic algorithm, and therefore, the scale of population will be according to tool
The problem of body, chooses suitable quantity, and the scale of initial population of the present invention is 30.
This paper fitness function is set as the inverse of neural network error sum of squares:
Wherein, SE is the error sum of squares between the prediction output of neural network and desired output.It can from fitness function
Know, neural network prediction error is smaller, and corresponding fitness function is bigger, and adaptability is better.
The selection of individual
The selection of individual can be carried out by probability value, and formula is as follows:
In formula, fiFor the fitness value of individual i；N is population at individual number.
Optimum individual does not have to carry out crossover operation, but is copied directly into the next generation.For other individuals, then use
Crossover probability pc.Crossover operation is carried out to 2 individuals, generates other 2 new individuals.Equally, optimum individual does not also carry out
Mutation operation, but it is copied directly to the next generation.It is then with mutation probability p for other individualsmMutation operation is carried out, is produced
Bear in addition new individual.In this experiment, pc=0.2, pm=0.1, evolutionary generation 40.
(3) training process of BP neural network
The weight and threshold value obtained using step (234) is trained BP neural network, until determining obtained error
Less than there is preset error value.
Best initial weights and threshold value that genetic algorithm obtains are used in BP neural network, the output of BP neural network is obtained,
Relative error is calculated, if relative error reaches the standard of setting, emulation obtains prediction result；If the error mark of not up to setting
Standard then returns and obtains best initial weights and threshold value, until the best initial weights and threshold value of acquisition, exports result and relative error is calculated
Until reaching established standards.
Then the BP neural network after will become trained at function is used as the prediction model of prediction underground water buried depth, i.e., based on collection
It is divided into the determination of Set Pair Analysis optimal independent variables, BP neural network structure to analysis and genetic optimization BP neural network prediction model
With three parts of genetic algorithm optimization BP neural network, specific flow chart is as figure 5 illustrates.
Wherein, the BP neural network structure determination based on Set Pair Analysis preferably inputs parameter according to Set-pair Analysis Model
Number determines BP neural network structure, and then determines the length of genetic algorithm individual.There are 10 input parameters in the present invention, 1
Output parameter, BP neural network structure are 10-19-1, i.e. input layer has 10 nodes, and hidden layer has 19 nodes, and output layer has
1 node shares 209 weights, 20 threshold values, so genetic algorithm individual UVR exposure length is 229.From 439 groups of input and output
343 groups are selected in data sequentially in time as training data, is used for network training, 96 groups are used as test data.Training
Data predict Error Absolute Value and as ideal adaptation angle value, and ideal adaptation angle value is smaller, and the individual is more excellent.
BP neural network termination condition setting are as follows: frequency of training 5000, convergence error 0.001；Genetic algorithm parameter setting
Are as follows: evolution number is 40 times；Population scale n is difficult to find out optimal solution when too small, and too big then convergence time is longer, general n=20
~80, taking population scale is 30, crossover probability PcIt is 0.2；Mutation probability PmIt is 0.1.Genetic algorithm optimization obtains BP neural network
Optimal initial weight and threshold value are assigned to BP neural network by optimal initial weight and threshold value, such as table 1, are trained with training data
Nonlinear function output is predicted after 343 times, prediction result is as shown in Figure 4.
Table 1
As seen from Figure 6, the prediction based on Set Pair Analysis and genetic algorithm optimization BP neural network to underground water buried depth
Effect is best, and the relative average error that prediction is obtained by calculation is 0.0884, it can be found that being calculated based on Set Pair Analysis and heredity
Method Optimized BP Neural Network model is smaller than genetic algorithm optimization BP neural network model predictive error, and precision of prediction is higher, should
The error of model prediction output and desired output is as shown in Figure 7.
The independent variable that the present invention is preferably gone out with Set-pair Analysis Model applies BP neural network and genetic algorithm optimization BP mind
The prediction of underground water buried depth has been carried out through network and based on Set Pair Analysis and genetic optimization BP neural network prediction model.Will more than
Obtained prediction result is compared, and is listed as shown in table 2:
Table 2
As can be seen from Table 1, based on Set Pair Analysis and genetic optimization BP neural network model predictive error compared with genetic optimization
BP neural network and BP neural network model predictive error are smaller, and the building process reliability of three kinds of methods is found from prediction result
It is higher.
Conclusion: the present invention using 10 independents variable that Set Pair Analysis preferably goes out as input, is built respectively using BP neural network
Underground water buried depth prediction model has been found, and it has been optimized, has been established based on Set Pair Analysis and genetic algorithm optimization BP nerve
The prediction model of the underground water buried depth of network.On the basis of BP neural network algorithm and genetic algorithm idea, structure is established
For the neural network of 10-19-1, by verification experimental verification as can be seen that the relative average error of BP neural network training stage is
0.1210, the relative average error 0.1508 of forecast period；The BP Neural Network of Genetic Algorithms training rank of independent variable screening is not done
The relative average error of section is 0.0867, and such as Fig. 4, the relative average error of forecast period is 0.0943；Based on Set Pair Analysis and
The relative average error 0.0798 of genetic algorithm optimization BP neural network training stage, as shown in Figure 6 and Figure 7, forecast period
Relative average error 0.0884.Three kinds of model results show: BP neural network is general to the effect of the prediction of underground water buried depth, loses
Propagation algorithm Optimized BP Neural Network is preferable than BP neural network to the prediction of underground water buried depth, excellent based on Set Pair Analysis and genetic algorithm
The BP Neural Network of Genetic Algorithms prediction that change BP neural network less does independent variable screening is more stable, meets actual requirement, is
Underground water buried depth provides effective solution method.
The above-described embodiments merely illustrate the principles and effects of the present invention, and is not intended to limit the present invention.It is any ripe
The personage for knowing this technology all without departing from the spirit and scope of the present invention, carries out modifications and changes to above-described embodiment.Cause
This, institute is complete without departing from the spirit and technical ideas disclosed in the present invention by those of ordinary skill in the art such as
At all equivalent modifications or change, should be covered by the claims of the present invention.
Claims (7)
1. a kind of prediction technique based on genetic algorithm optimization BP neural network underground water buried depth, which is characterized in that the method
It includes at least:
(1) acquisition of data
It is selected with the regional higher independent variable of underground water buried depth Pair Analysis to be measured using Method of Set Pair Analysis as target independent variable,
And obtain the corresponding sample of the target independent variable；
(1) setting of BP neural network
Using the corresponding sample of the target independent variable as the training sample of BP neural network, the input section of BP neural network is determined
The quantity of point, output node and hidden layer obtains the output result of BP neural network；
(2) genetic Optimization Algorithm
(21) genetic algorithm initial parameter is set
Population scale, evolution number, crossover probability and mutation probability are set, the weight of BP neural network and threshold value are formed into heredity
Coding；
(22) building of fitness function
Using the function that BP neural network prediction output is formed with desired output as fitness value；
(23) genetic algorithm treatment process
(231) individual choice operates: being used to generate the next generation, selection principle with predetermined probabilities selection individual in old group are as follows:
Fitness value according to individual is selected from big to small；
(232) crossover operation: refer to the foundation for selecting a plurality of parent chromosome to carry out next-generation chromosome from population, pass through dye
The combined crosswise of colour solid generates new individual；
(233) mutation operation: referring to an optional individual from group, makes a variation to certain section of coding in selected chromosome
To generate new individual；
(234) repeat step (231) to step (233), constantly individual in population is selected, is intersected, mutation operation simultaneously
Fitness value is recorded, reaches evolution maximum number of times, the genetic coding in the individual newly obtained is decoded, calculates fitness,
Group is compared with original seed；The corresponding chromosome of fitness optimal solution is threshold value corresponding to BP neural network and weight；
(3) training process of BP neural network
The weight and threshold value obtained using step (234) is trained BP neural network, until determining that obtained error is less than
There is preset error value.
2. the prediction technique according to claim 1 based on genetic algorithm optimization BP neural network underground water buried depth, special
Sign is that the BP neural network termination condition is arranged are as follows:
Frequency of training 5000, convergence error 0.001, input layer have 10 nodes, and hidden layer has 19 nodes, and output layer has 1
Node.
3. the prediction technique according to claim 1 or 2 based on genetic algorithm optimization BP neural network underground water buried depth,
It is characterized in that, the genetic algorithm parameter setting are as follows:
Evolution number is 40 times；Population scale n value range is 20~80, and taking population scale is 30, crossover probability 0.2；Become
Different probability is 0.1.
4. the prediction technique according to claim 1 or 2 based on genetic algorithm optimization BP neural network underground water buried depth,
It is characterized in that, it is described to be selected with the higher independent variable of areal association degree to be measured using Method of Set Pair Analysis as target independent variable
The specific formula that step is taken are as follows:
Setting first set A indicates that independent variable, second set B indicate the measured value in area to be measured, and the collection of first set A and second
Conjunction B composition integrates to as H:
H=(A, B)
Wherein, A and B has N number of characteristic features, comprising: S denominator, P opposed characteristics, F difference characteristic.If enabling a
=S/N, b=F/N, c=P/N, then collecting indicates the Pair Analysis (frequently referred to three-unit connection number) of H are as follows:
U=a+bI+cJ
I=a/ (a+c)+c/ (a+c) J
Wherein, U is Pair Analysis；A is unification degree；B is diversity factor；C is opposition degree；And a+b+c=1；I is poor
Heterolabeling symbol, I ∈ [- 1,1]；J is opposition degree coefficient, and value is -1.
5. the prediction technique according to claim 1 or 2 based on genetic algorithm optimization BP neural network underground water buried depth,
It is characterized in that, the code length of the genetic algorithm are as follows:
S=n × m+m × l+l
In formula, m is node in hidden layer；N is input layer number；L output layer number of nodes.
6. the prediction technique according to claim 1 based on genetic algorithm optimization BP neural network underground water buried depth, special
Sign is that the fitness function embodies are as follows:
In formula, SE is the error sum of squares between the prediction output of neural network and desired output.
7. the prediction technique according to claim 1 based on genetic algorithm optimization BP neural network underground water buried depth, special
Sign is that individual select probability value embodies are as follows:
In formula, PiFor the select probability value of individual i, fiFor the fitness value of individual i；K is population at individual number.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811092233.3A CN109409568A (en) | 2018-09-19 | 2018-09-19 | A kind of prediction technique based on genetic algorithm optimization BP neural network underground water buried depth |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811092233.3A CN109409568A (en) | 2018-09-19 | 2018-09-19 | A kind of prediction technique based on genetic algorithm optimization BP neural network underground water buried depth |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109409568A true CN109409568A (en) | 2019-03-01 |
Family
ID=65465099
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811092233.3A Pending CN109409568A (en) | 2018-09-19 | 2018-09-19 | A kind of prediction technique based on genetic algorithm optimization BP neural network underground water buried depth |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109409568A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110059821A (en) * | 2019-03-19 | 2019-07-26 | 埃睿迪信息技术（北京）有限公司 | Neural network optimization, device, computer equipment and storage medium |
| CN110288726A (en) * | 2019-06-28 | 2019-09-27 | 广西民族大学 | One kind being based on BP neural network sugarcane transfer car(buggy) risk prediction method |
| CN110457758A (en) * | 2019-07-16 | 2019-11-15 | 江西理工大学 | Prediction technique, device, system and the storage medium in Instability of Rock Body stage |
| CN110796234A (en) * | 2019-10-21 | 2020-02-14 | 苏州浪潮智能科技有限公司 | Method and device for predicting computer state |
| CN110909413A (en) * | 2019-12-09 | 2020-03-24 | 山东大学 | Concrete panel rock-fill dam horizontal displacement prediction accuracy method and system |
| CN111028512A (en) * | 2019-12-31 | 2020-04-17 | 福建工程学院 | Real-time traffic prediction method and device based on sparse BP neural network |
| CN112247673A (en) * | 2020-09-07 | 2021-01-22 | 博深普锐高（上海）工具有限公司 | Woodworker cutter wear state diagnosis method based on genetic BP neural network |
| CN113077041A (en) * | 2020-01-06 | 2021-07-06 | 大唐环境产业集团股份有限公司 | Deep neural network modeling independent variable dimension reduction method based on genetic algorithm |
| CN113971487A (en) * | 2021-10-25 | 2022-01-25 | 武汉船舶通信研究所（中国船舶重工集团公司第七二二研究所） | Method, system, terminal and medium for predicting water depth |
| RU2784749C2 (en) * | 2021-04-21 | 2022-11-29 | Федеральное государственное бюджетное образовательное учреждение высшего образования "Кубанский государственный технологический университет" (ФГБОУ ВО "КубГТУ") | Modified controller for control of training neural network with genetic algorithm |
| US11628445B2 (en) | 2019-04-09 | 2023-04-18 | Jiangxi University Of Science And Technology | Material crushing cavity structure and method for designing a multi-stage nested material crushing cavity structure |
| US11989655B2 (en) | 2020-10-26 | 2024-05-21 | Jiangxi University Of Science And Technology | Prediction method, device and system for rock mass instability stages |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102073952A (en) * | 2011-03-07 | 2011-05-25 | 北京师范大学 | Water resource bearing capacity evaluation method |
| CN102789445A (en) * | 2012-07-13 | 2012-11-21 | 南京大学 | Establishment method for wavelet analysis and rank set pair analysis of medium and long-term hydrological forecast model |
| CN105243436A (en) * | 2015-09-19 | 2016-01-13 | 东北电力大学 | Wind power real-time prediction method based on rank set pair analysis |
| CN106845142A (en) * | 2017-03-09 | 2017-06-13 | 南京大学 | Quality evaluation method based on improved rough set Set Pair Analysis |
| CN107464045A (en) * | 2017-07-21 | 2017-12-12 | 河海大学 | A kind of modernization of water resources indication system judgment method based on cloud model |
-
2018
- 2018-09-19 CN CN201811092233.3A patent/CN109409568A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102073952A (en) * | 2011-03-07 | 2011-05-25 | 北京师范大学 | Water resource bearing capacity evaluation method |
| CN102789445A (en) * | 2012-07-13 | 2012-11-21 | 南京大学 | Establishment method for wavelet analysis and rank set pair analysis of medium and long-term hydrological forecast model |
| CN105243436A (en) * | 2015-09-19 | 2016-01-13 | 东北电力大学 | Wind power real-time prediction method based on rank set pair analysis |
| CN106845142A (en) * | 2017-03-09 | 2017-06-13 | 南京大学 | Quality evaluation method based on improved rough set Set Pair Analysis |
| CN107464045A (en) * | 2017-07-21 | 2017-12-12 | 河海大学 | A kind of modernization of water resources indication system judgment method based on cloud model |
Non-Patent Citations (3)
| Title |
|---|
| CHENGGUO WU, ETC: "Set Pair Analysis and BP Neural Network Coupling Model for Optimal Selection of Flood Control Operation", 《2010 CHINESE CONTROL AND DECISION CONFERENCE》 * |
| 赵晓慎 等: "集对分析在改进BP神经网络凌汛开河日期预测评估中的应用", 《水电能源科学》 * |
| 陈笑 等: "基于遗传算法的 BP 神经网络模型在地下水埋深预测中的应用——以蒙城县为例", 《水利水电技术》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110059821A (en) * | 2019-03-19 | 2019-07-26 | 埃睿迪信息技术（北京）有限公司 | Neural network optimization, device, computer equipment and storage medium |
| US11628445B2 (en) | 2019-04-09 | 2023-04-18 | Jiangxi University Of Science And Technology | Material crushing cavity structure and method for designing a multi-stage nested material crushing cavity structure |
| CN110288726A (en) * | 2019-06-28 | 2019-09-27 | 广西民族大学 | One kind being based on BP neural network sugarcane transfer car(buggy) risk prediction method |
| CN110457758A (en) * | 2019-07-16 | 2019-11-15 | 江西理工大学 | Prediction technique, device, system and the storage medium in Instability of Rock Body stage |
| CN110796234B (en) * | 2019-10-21 | 2022-07-12 | 苏州浪潮智能科技有限公司 | Method and device for predicting computer state |
| CN110796234A (en) * | 2019-10-21 | 2020-02-14 | 苏州浪潮智能科技有限公司 | Method and device for predicting computer state |
| CN110909413A (en) * | 2019-12-09 | 2020-03-24 | 山东大学 | Concrete panel rock-fill dam horizontal displacement prediction accuracy method and system |
| CN111028512A (en) * | 2019-12-31 | 2020-04-17 | 福建工程学院 | Real-time traffic prediction method and device based on sparse BP neural network |
| CN113077041A (en) * | 2020-01-06 | 2021-07-06 | 大唐环境产业集团股份有限公司 | Deep neural network modeling independent variable dimension reduction method based on genetic algorithm |
| CN112247673A (en) * | 2020-09-07 | 2021-01-22 | 博深普锐高（上海）工具有限公司 | Woodworker cutter wear state diagnosis method based on genetic BP neural network |
| US11989655B2 (en) | 2020-10-26 | 2024-05-21 | Jiangxi University Of Science And Technology | Prediction method, device and system for rock mass instability stages |
| RU2784749C2 (en) * | 2021-04-21 | 2022-11-29 | Федеральное государственное бюджетное образовательное учреждение высшего образования "Кубанский государственный технологический университет" (ФГБОУ ВО "КубГТУ") | Modified controller for control of training neural network with genetic algorithm |
| CN113971487A (en) * | 2021-10-25 | 2022-01-25 | 武汉船舶通信研究所（中国船舶重工集团公司第七二二研究所） | Method, system, terminal and medium for predicting water depth |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109409568A (en) | A kind of prediction technique based on genetic algorithm optimization BP neural network underground water buried depth | |
| CN111310968B (en) | LSTM neural network circulating hydrologic forecasting method based on mutual information | |
| CN110647900B (en) | Intelligent safety situation prediction method, device and system based on deep neural network | |
| CN109142171B (en) | Urban PM10 concentration prediction method based on feature expansion and fusing with neural network | |
| CN108009674A (en) | Air PM2.5 concentration prediction methods based on CNN and LSTM fused neural networks | |
| CN109902801A (en) | A kind of flood DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM method based on variation reasoning Bayesian neural network | |
| CN104239489B (en) | Utilize the method for similarity searching and improved BP forecast level | |
| CN109285346A (en) | A kind of city road net traffic state prediction technique based on key road segment | |
| CN111160628A (en) | Air pollutant concentration prediction method based on CNN and double-attention seq2seq | |
| CN110335270A (en) | Transmission line of electricity defect inspection method based on the study of hierarchical regions Fusion Features | |
| CN109272146A (en) | A kind of Forecasting Flood method corrected based on deep learning model and BP neural network | |
| CN111915058B (en) | Flood prediction method and device based on long-time memory network and transfer learning | |
| CN113887787B (en) | Flood forecast model parameter multi-objective optimization method based on long-short-term memory network and NSGA-II algorithm | |
| Dumedah et al. | Selecting model parameter sets from a trade-off surface generated from the non-dominated sorting genetic algorithm-II | |
| CN110738355A (en) | urban waterlogging prediction method based on neural network | |
| CN109840587A (en) | Reservoir reservoir inflow prediction technique based on deep learning | |
| CN108460461A (en) | Mars earth shear parameters prediction technique based on GA-BP neural networks | |
| CN111343650A (en) | Urban scale wireless service flow prediction method based on cross-domain data and loss resistance | |
| CN111311001B (en) | Bi-LSTM network short-term load prediction method based on DBSCAN algorithm and feature selection | |
| CN115544239A (en) | Deep learning model-based layout preference prediction method | |
| CN114584406B (en) | Industrial big data privacy protection system and method for federated learning | |
| CN115759445A (en) | Machine learning and cloud model-based classified flood random forecasting method | |
| CN115329930A (en) | Flood process probability forecasting method based on mixed deep learning model | |
| Yadav et al. | Application of Artificial Neural Network and Genetic Algorithm Based Artificial Neural Network Models for River Flow Prediction. | |
| CN115051925A (en) | Time-space sequence prediction method based on transfer learning |
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: 20190301 |
|
| RJ01 | Rejection of invention patent application after publication |