CN107729671A - A kind of mine laneway coefficient of frictional resistance Forecasting Methodology based on SVMs - Google Patents
A kind of mine laneway coefficient of frictional resistance Forecasting Methodology based on SVMs Download PDFInfo
- Publication number
- CN107729671A CN107729671A CN201711032948.5A CN201711032948A CN107729671A CN 107729671 A CN107729671 A CN 107729671A CN 201711032948 A CN201711032948 A CN 201711032948A CN 107729671 A CN107729671 A CN 107729671A
- Authority
- CN
- China
- Prior art keywords
- coefficient
- frictional resistance
- parameter
- prediction
- model
- 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]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/13—Architectural design, e.g. computer-aided architectural design [CAAD] related to design of buildings, bridges, landscapes, production plants or roads
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/08—Probabilistic or stochastic CAD
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Geometry (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- Computer Hardware Design (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Architecture (AREA)
- Genetics & Genomics (AREA)
- Physiology (AREA)
- Civil Engineering (AREA)
- Structural Engineering (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of mine laneway coefficient of frictional resistance Forecasting Methodology based on SVMs, LIBSVM is used under MATLAB environment, using cold water well colliery bolting and shotcrete roadway, roadway supported by bolt, bolt-mesh-anchor support tunnel coefficient of frictional resistance as research object, training set is trained with trellis search method, genetic algorithms approach, particle cluster algorithm method respectively, obtain optimal Radial basis kernel function parameter and punishment parameter combination, forecast model is established using the parameter combination of selection, checking collection is predicted.Prediction result shows that the forecast model that the parameter combination that these three methods are chosen is established can effectively predict mine laneway coefficient of frictional resistance, demonstrate the feasibility of SVM prediction mine laneway coefficient of frictional resistance.Wherein, the model prediction result combination property of genetic algorithm Selecting All Parameters combination behaves oneself best, and has carried out proving and comparisom with mine measured result.
Description
Technical field
Mine laneway coefficient of frictional resistance of the present invention predicts field, and in particular to a kind of based on SVMs
Mine laneway coefficient of frictional resistance Forecasting Methodology.
Background technology
In Safety of Coal Mine Production, Safety first, precaution crucial, and the work of " one through three prevention " has important meaning to colliery
Justice.Mine ventilation is auxiliary link important in coal mine production system, and it is that ozone is transported into underground from ground,
There is provided for the operating personnel of underground and create good working environment, while also dilute and exclude the various poisonous, You Haiqi in underground
Body and mine dust, the temperature and humidity of underground are adjusted, improve underground weather conditions.The quality of mine characteristic directly affects
Mine safety production and its economic benefit.Therefore, to ensure Safety of Coal Mine Production, research mine characteristic has non-
Often important realistic meaning.
Mine laneway coefficient of frictional resistance is one of reflection most important technical parameter of mine laneway drag characteristic.It is one
Individual constant value, reflect the intrinsic essential attribute in tunnel.It is Design of Mine Ventilation, improvement on Ventilation System and strengthened logical simultaneously
The technical foundation of the work such as wind technical management.
The determination of mine laneway coefficient of frictional resistance, conventional reality, which measures and tabled look-up, chooses two methods.Actually measure
Tunnel coefficient of frictional resistance value can reflect the drag characteristic of mine laneway.But measuring environment is poor, workload is big and need compared with
More survey crews, co-ordination is difficult, and the normal production work of mine is necessarily affected in measurement process.Look-up table is simply easily grasped
Make, but tunnel coefficient of frictional resistance table the last century 80's formulation used by China's Design of Mine Ventilation at present.Due to
Mine geomorphic feature, the occurrence condition of coal are different, cause mine laneway coefficient of frictional resistance value different.Even same ore deposit is same
The tunnel of kind Support types, also because of the influence of the section in tunnel, length etc., there is also very big difference for mine laneway coefficient of frictional resistance
It is different.China's coal-mine industry has very big by the development, its well type, scale, roadway layout, supporting construction etc. of many decades
Change, existing tunnel coefficient of frictional resistance table does not grow with each passing hour, is updated in time, it is impossible to adapts to present reality completely
Situation.Mine laneway coefficient of frictional resistance numerical value is smaller, tends not to the enough attention for enough causing mine ventilation technology personnel,
Before tunnel coefficient of frictional resistance is chosen, well aeration technology personnel combine the specific coal seam of each mine and assigned with not suiting measures to local conditions
The actual conditions for depositing condition, roadway layout situation, Support types and Ventilation Structures are chosen.Choosing mine laneway friction
During resistance coefficient, estimation of simply simply tabling look-up, value has larger randomness, therefore human factor can produce to result of calculation
Influence, even result in and actual result deviation is larger.
With this computer science and statistical development, computer learning has become what every field was predicted
A kind of common method.Machine learning is built upon on the basis of data, after obtaining or finding certain rule, then is reversely returned
To be illustrated to data, explaination, and also its effect is also manifested by the deduction and prediction to Future Data development trend, that is, promotes
Ability.SVMs is wherein most typical one kind, is adapted to small sample problem, and its foundation experienced many decades hair so far
Exhibition is able to ripe utilization.Mining industry field at home, applying for SVMs has begun in the beginning of this century, by the more than ten years
Development, it has been used for carrying out water bursting in mine risk profile, water sources in coal mines analysis, mine fire risk profile, coal
Country rock Prediction for Rock Burst, coal petrography image classification identification technology are exploited, therefore the technology is applied to mine laneway coefficient of frictional resistance
Research is by with good prospect.
The content of the invention
To solve the above problems, the invention provides a kind of mine laneway coefficient of frictional resistance based on SVMs is pre-
Survey method.
To achieve the above object, the technical scheme taken of the present invention is:
A kind of mine laneway coefficient of frictional resistance Forecasting Methodology based on SVMs, comprises the following steps:
S1, the input/output variable for determining SVM prediction mine laneway coefficient of frictional resistance, real data is gathered,
And real data is normalized, obtain the training sample data for forecast model;
S2, choose core letter of the Radial basis kernel function as SVM prediction mine laneway coefficient of frictional resistance model
Number;
S3, using genetic algorithm the hyper parameter of forecast model is determined, and whole training set is trained, obtained
The model of SVM prediction mine laneway coefficient of frictional resistance;
S4, the data concentrated using the forecast model that step S3 is established to checking are predicted.
Wherein, the genetic algorithm is chosen SVM prediction model parameter combination and completed by following steps:
S31, the coding of chromosome, decoding and the initialization of population
Using the coded system based on real number, the punishment parameter C and kernel functional parameter g of SVM prediction model are set
Hunting zone, then the parameter combination value with feasibility in setting range is converted to the dye that can be received by genetic algorithm
Colour solid, n are finally randomly selected in initial data and carries out initialization process with to completing population;
S32, determine population at individual fitness
Using the mean square error of forecast model in the training process as individual adaptation degree evaluation index, i.e., some contaminates in population
Superiority-inferiority of the colour solid relative to other chromosomes during evolution;
S33, the selection of chromosome, intersection and variation
Individual is selected according to championship principle, after the select probability for giving individual, can be produced equal between [0,1]
Even random number, these random numbers can determine the mating probability of individual, participate in the individual that vitality is strong in population, fitness is high new
Colony is generated, while the individual in original colony is selected the superior and eliminated the inferior, and makes the equal iteration of fitness value of each individual in colony
Close to globally optimal solution.
The invention has the advantages that:
LIBSVM is used under MATLAB environment, with cold water well colliery bolting and shotcrete roadway, roadway supported by bolt, bolt-mesh-anchor support
The coefficient of frictional resistance in tunnel is research object, respectively with trellis search method, genetic algorithms approach, particle cluster algorithm method pair
Training set is trained, and is obtained optimal Radial basis kernel function parameter and punishment parameter combination, is built jointly using the parameter group of selection
Vertical forecast model, checking collection is predicted.Prediction result shows, the prediction mould that the parameter combination that these three methods are chosen is established
Type can effectively predict mine laneway coefficient of frictional resistance, demonstrate SVM prediction mine laneway coefficient of frictional resistance
Feasibility.Wherein, the model prediction result combination property of genetic algorithm Selecting All Parameters combination behaves oneself best, and surveys and tie with mine
Fruit has carried out proving and comparisom.
Brief description of the drawings
Fig. 1 SVM prediction flow charts.
The training sample of Fig. 2 bolting and shotcrete roadway coefficient of frictional resistance forecast models.
The checking sample of Fig. 3 bolting and shotcrete roadway coefficient of frictional resistance forecast models.
The training sample of Fig. 4 roadway supported by bolt coefficient of frictional resistance forecast models.
The checking sample of Fig. 5 roadway supported by bolt coefficient of frictional resistance forecast models.
The training sample of Fig. 6 bolt-mesh-anchor supports tunnel coefficient of frictional resistance forecast model.
The checking sample of Fig. 7 bolt-mesh-anchor supports tunnel coefficient of frictional resistance forecast model.
Fig. 8 grid data services choose SVM prediction bolting and shotcrete roadway coefficient of frictional resistance model parameter combined result
3D views.
Fig. 9 genetic algorithms choose the result of SVM prediction bolting and shotcrete roadway coefficient of frictional resistance model parameter combination.
Figure 10 particle cluster algorithms choose the knot of SVM prediction bolting and shotcrete roadway coefficient of frictional resistance model parameter combination
Fruit.
Figure 11 grid data services choose the combination of SVM prediction roadway supported by bolt coefficient of frictional resistance model parameter
As a result 3D views.
Figure 12 genetic algorithms choose the combination of SVM prediction roadway supported by bolt coefficient of frictional resistance model parameter
As a result.
Figure 13 particle cluster algorithms choose the combination of SVM prediction roadway supported by bolt coefficient of frictional resistance model parameter
Result.
Figure 14 grid data services choose SVM prediction bolt-mesh-anchor support tunnel coefficient of frictional resistance model parameter group
Close the 3D views of result.
Figure 15 genetic algorithms choose the coefficient of frictional resistance model parameter combination of SVM prediction bolt-mesh-anchor support tunnel
Result.
Figure 16 particle cluster algorithms choose SVM prediction bolt-mesh-anchor support tunnel coefficient of frictional resistance model parameter group
The result of conjunction.
The bolting and shotcrete roadway coefficient of frictional resistance value of Figure 17 grid data services Selecting All Parameters prediction and the contrast of measured value.
The bolting and shotcrete roadway coefficient of frictional resistance value of Figure 18 grid data services Selecting All Parameters prediction and the relative error of measured value.
The bolting and shotcrete roadway coefficient of frictional resistance value of Figure 19 genetic algorithms Selecting All Parameters prediction and the contrast of measured value.
The bolting and shotcrete roadway coefficient of frictional resistance value of Figure 20 genetic algorithms Selecting All Parameters prediction and the relative error of measured value.
The bolting and shotcrete roadway coefficient of frictional resistance of Figure 21 particle cluster algorithms Selecting All Parameters prediction contrasts with measured value
The bolting and shotcrete roadway coefficient of frictional resistance value of Figure 22 particle cluster algorithms Selecting All Parameters prediction and the relative error of measured value.
The roadway supported by bolt coefficient of frictional resistance value of Figure 23 grid data services Selecting All Parameters prediction and the contrast of measured value.
Figure 24 grid data services Selecting All Parameters prediction roadway supported by bolt coefficient of frictional resistance value and measured value it is relative
Error.
The roadway supported by bolt coefficient of frictional resistance value of Figure 25 genetic algorithms Selecting All Parameters prediction and the contrast of measured value.
The roadway supported by bolt coefficient of frictional resistance value of Figure 26 genetic algorithms Selecting All Parameters prediction and the relative of measured value are missed
Difference.
The roadway supported by bolt coefficient of frictional resistance value of Figure 27 particle cluster algorithms Selecting All Parameters prediction and the contrast of measured value.
Figure 28 particle cluster algorithms Selecting All Parameters prediction roadway supported by bolt coefficient of frictional resistance value and measured value it is relative
Error.
The bolt-mesh-anchor support tunnel coefficient of frictional resistance value of Figure 29 grid data services Selecting All Parameters prediction and pair of measured value
Than.
The bolt-mesh-anchor support tunnel coefficient of frictional resistance value and the phase of measured value of Figure 30 grid data services Selecting All Parameters prediction
To error.
The anchor rete cord tunnel coefficient of frictional resistance value of Figure 31 genetic algorithms Selecting All Parameters prediction and the contrast of measured value.
The anchor rete cord tunnel coefficient of frictional resistance value of Figure 32 genetic algorithms Selecting All Parameters prediction and the relative error of measured value.
The contrast of the anchor rete cord tunnel coefficient of frictional resistance and measured value of the prediction of Figure 33 particle cluster algorithms Selecting All Parameters.
The anchor rete cord tunnel coefficient of frictional resistance value of Figure 34 particle cluster algorithms Selecting All Parameters prediction and the relative of measured value are missed
Difference.
Embodiment
In order that objects and advantages of the present invention are more clearly understood, the present invention is carried out with reference to embodiments further
Describe in detail.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not used to limit this hair
It is bright.
Embodiment
SVM prediction flow chart is as shown in figure 1, comprise the following steps:
1. the selection of training set
The input/output variable of SVM prediction mine laneway coefficient of frictional resistance is determined, gathers real data, and
Real data is normalized, obtains the training sample data for forecast model.
2. the selection of kernel function
Choose kernel function of the Radial basis kernel function as SVM prediction mine laneway coefficient of frictional resistance model.
3. the selection of model parameter
Grid data service, genetic algorithm and particle cluster algorithm is respectively adopted to be determined the hyper parameter of forecast model, and
Whole training set is trained, obtains the model of SVM prediction mine laneway coefficient of frictional resistance.
4. the foundation of forecast model
Using the forecast model of 3. middle foundation, the data concentrated to checking are predicted, and experimental result is analyzed;
According to prediction effect, most suitable forecast model is determined.
Forecast model evaluation index
Introduce following 2 evaluation and foreca degree of fitting indexs.
(1) relative error (Relative Percentage Error, RFE)
(2) Mean Square Error (Mean Square Error, MSE), is defined as the difference of prediction result and actual value
Square arithmetic average:
Wherein, X1Represent actual value,Represent predicted value, n representative sample numbers.
Mean Square Error can preferably reflect the precision of prediction, but be not easy to weigh unbiasedness.Relative error can be very well
Measurement forecast model unbiasedness, coordinate Mean Square Error use can preferably complementation, can be to predicted value and actual value
Fitting degree carry out effective evaluation.
Forecast model samples selection and pretreatment
(1) cross validation
Initial data is divided into by k groups using k folding cross-validation methods (k-fold Cross Validation, k-CV), will
Each of subset data makees checking collection respectively, and remaining k-1 groups data show that k can reflect that training set returns as training set
The parameter of the degree of accuracy, by the use of mean square deviation average as this k-CV under evaluation return performance index.
(2) selection of training sample
1. bolting and shotcrete roadway coefficient of frictional resistance sample data is 130 groups in the present embodiment, with reference to the thought of cross validation, sheet
Embodiment uses 5 folding cross-validation methods, and 80 groups of data (referring to table 1) by before, as shown in Fig. 2 as SVM prediction
The training sample of bolting and shotcrete roadway coefficient of frictional resistance model;Remaining 50 groups (referring to table 1), as shown in figure 3, as supporting vector
Machine predicts the checking sample of bolting and shotcrete roadway coefficient of frictional resistance.
2. roadway supported by bolt coefficient of frictional resistance sample data is 45 groups in the present embodiment, with reference to the think of of cross validation
Think, the present embodiment uses 5 folding cross-validation methods, and 30 groups of data (referring to table 2) by before, as shown in figure 4, as SVMs
Predict the training sample of roadway supported by bolt coefficient of frictional resistance model;Remaining 15 groups (referring to table 2), as shown in figure 5, conduct
The checking sample of SVM prediction roadway supported by bolt coefficient of frictional resistance model.
3. bolt-mesh-anchor support tunnel coefficient of frictional resistance sample data is 50 groups in the present embodiment, with reference to the think of of cross validation
Think, the present embodiment uses 5 folding cross-validation methods, and 30 groups of data (referring to table 3) by before, as shown in fig. 6, as SVMs
Predict the training sample of bolt-mesh-anchor support tunnel coefficient of frictional resistance model;Remaining 20 groups (referring to table 3), as shown in fig. 7, making
To hold the checking sample that vector machine predicts bolt-mesh-anchor support tunnel coefficient of frictional resistance model.
The bolting and shotcrete roadway coefficient of frictional resistance value of table 1
The roadway supported by bolt coefficient of frictional resistance value of table 2
The bolt-mesh-anchor support tunnel coefficient of frictional resistance value of table 3
Sample data pre-processes
Data are normalized:
Wherein, xiNormalized initial data, x are treated in expressionminAnd xmaxFor the minimax value of initial data, y is normalizing
Data after change, its span are [a, b].
To cause flooding for data message when avoiding cross-validation method processing data, data are normalized the present embodiment
Pretreatment, by data normalization to section [0,1].
The choosing method of parameter
Grid data service
In SVM prediction model, the selection of parameter combination is exactly first to set punishment parameter C and RBF
Parameter g selection range, certain step-length is then set, all possible parameter combination is taken using grid data service
Value, for taking fixed C and g to obtain the training checking collection forecast set under this group of C and g using k-CV methods, finally take so that verifying
Collect predictablity rate highest that group of C and g as optimal parameter.
The optimal parameter group of SVM prediction mine laneway coefficient of frictional resistance model is found using grid data service
When closing C and g, the parameter optimization function in LIBSVM tool boxes can be utilized.
Grid data service parameter combination optimizing function (regression problem):SVMcgForRegress
[bestCVmse, bestc, bestg]=SVMcgForRegress (train_label, train, cmin, cmax,
Gmin, gmax, v, cstep, gstep, msestep)
(1) input:
train:Training set, the same svmtrain of form.
train_label:The label of training set, the same svmtrain of form.
Gmin, gmax:Radial basis kernel function parameter g span, i.e. g are found in the range of [2^gmin, 2^gmax]
Optimal Radial basis kernel function parameter g, default value gmin=-8, gmax=8,
The scope for giving tacit consent to g is [2^ (- 8), 2^8].
Cmin, cmax:Punishment parameter C span, i.e., optimal parameter is found in the range of [2^cmin, 2^cmax]
C,
Default value is cmin=-8, cmax=8, that is, the scope for giving tacit consent to punishment parameter C is [2^ (- 8), 2^8].
v:Parameters in cross validation, i.e., v folding cross validations are carried out to training sample, are defaulted as 3,
I.e. acquiescence carries out 3 folding crossover process.
The present embodiment is verified using 5 folding cross-validation methods to training sample.
accstep:In parameter combination selection result figure, stepped intervals size that accuracy rate discretization is shown,
Span is [0,100], is defaulted as 4.5.
Cstep, gstep:C and g step size when being predicted model parameter combination optimizing,
That is C value is 2^cmin, 2^ (cmin+cstep) ..., 2^cmax,
G value is 2^gmin, 2^ (gmin+gstep) ..., 2^gmax,
Acquiescence value is cstep=1, gstep=1.
(2) export:
bestg:Optimal parameter g.
bestC:Optimal parameter C.
bestCVmse:Final CV mean square deviation.
Genetic algorithm
It is as follows that genetic algorithm chooses the step of SVM prediction model parameter combination:
(1) coding of chromosome, decoding and the initialization of population
Using the coded system based on real number, the punishment parameter C and kernel functional parameter g of SVM prediction model are set
Hunting zone, then the parameter combination value with feasibility in setting range is converted to the dye that can be received by genetic algorithm
Colour solid, n are finally randomly selected in initial data and carries out initialization process with to completing population.
(2) population at individual fitness is determined
Population at individual fitness is superiority-inferiority of some chromosome relative to other chromosomes during evolution in population
Embodiment.Some chromosome fitness is higher, it is bigger to illustrate that it participates in the probability of genetic iteration next time, so that entirely
Population is evolved to global optimization.Mean square error of the present embodiment using forecast model in the training process is commented as individual adaptation degree
Valency index.
(3) selection of chromosome, intersection and variation
The present embodiment is selected individual according to championship principle, after the select probability for giving individual, can be produced [0,1]
Between uniform random number, these random numbers can determine the mating probability of individual, make that vitality is strong in population, fitness is high
Body participates in new colony's generation, while the individual in original colony is selected the superior and eliminated the inferior, and makes the fitness of each individual in colony
It is worth equal iteration close to globally optimal solution.
Crossover operator determines the ability of searching optimum of genetic algorithm.It is selected from initial population fitness it is larger two
Individual, some gene positions of the two individuals are exchanged at random, so as to produce the master data feature with parent
Newborn filial generation.Mutation operator determines the local search ability of genetic algorithm.It is simulate it is small general during natural evolution
The mutation of rate gene position, change the gene in individual chromosome coded strings by the less probability set in GA-SVM models
Position, so as to produce a new individual.
When being scanned for using genetic algorithm to SVM prediction model parameter combination, due in new caused colony
Include the group property of previous generation individuals, avoid the search to invalid parameter combination, thus genetic algorithm can be selected quickly
Take out the parameter combination for making SVM prediction model performance reach better effects.
The optimal parameter combination C of SVM prediction mine laneway coefficient of frictional resistance model is found using genetic algorithm
During with g, the parameter optimization function in LIBSVM tool boxes can be utilized.
Genetic algorithm parameter combination optimizing function (regression problem):gaSVMcgForClass
[bestCVmse, bestc, bestg, ga_option]=gaSVMcgForClass (train_label, train,
ga_option)
Ga_option=struct (' maxgen ', ' sizepop ', ' ggap ', ' cbound ', ' gbound ', ' v ');
Cbound=[cmin, cmax], parameter C span, be defaulted as (0,100];sizepop:
Population maximum quantity, 20 are defaulted as, general span is [20,100];
maxgen:Maximum evolutionary generation, 200 are defaulted as, general span is [100,500];The present embodiment takes 100.
Gbound=[gmin, gmax], parameter g span, is defaulted as [0,100].
The implication of other specification is identical with the parameter meaning in grid data service.
Particle cluster algorithm
It is as follows that particle cluster algorithm chooses the step of SVM prediction model parameter combination:
(1) population particle (i.e. the parameter combination of forecast model) position and speed are initialized;
(2) fitness calculating is carried out to each particle in population;
(3) the individual extreme value and global extremum of more new particle;
(4) speed of more new particle and position;
(5) judge whether particle fitness meets end condition, otherwise go to step (2).
Particle cluster algorithm method be it is a kind of parameter can selected value formed space in model parameter carry out parallel search
A kind of heuristic parameter selection method.Because it not only considers oneself combination of the model parameter through search, and scanning for
During can also refer to the influences of other particle search results in population, it is thus possible to quickly reach relatively good model ginseng
Number search effect.
The optimal parameter group of SVM prediction mine laneway coefficient of frictional resistance model is found using particle cluster algorithm
When closing C and g, the parameter optimization function in LIBSVM tool boxes can be utilized.
Utilize PSO parameter optimizations function (regression problem):psoSVMcgForRegress
[bestCVmse, bestc, bestg, pso_option]=
PsoSVMcgForRegress (train_label, train, pso_opt_ion);
Pso_option=struct (' c1 ', ' c2 ', ' maxgen ', ' sizepop ', ' k ', ' w
V ', ' wP ', ' v ', ' popcmax ',
' popcmin ', ' popgmax ', ' popgmin ');
maxgen:Maximum evolutionary generation, default value 100;
c1:Pso parameter local search abilities, initial value 1.5;
k:Speed and x relation (V=kx), k ∈ [0.1,1.0], initial value 0.6;
c2:Pso parameter global search capabilities, initial value 1.7;
sizepop:Population maximum quantity, default value 20;
wP:In population recruitment formula, the coefficient of elasticity before speed, 1 is initially;
wV:Before speed more new formula, the coefficient of elasticity wV ∈ [0.8,1.2] before speed, initial value 1;
popcmin:Parameter C minimum value, it is 0.1;
popcmax:Parameter C maximum occurrences, it is 100;
popgmin:Parameter g minimum value, it is 0.01.
popgmax:Parameter g maximum occurrences, it is 100;
The implication of other specification is identical with the parameter meaning in grid data service.
The selection range of parameter determines
It is Taiwan woods intelligence benevolence used by the present embodiment application SVM prediction mine laneway coefficient of frictional resistance
LIBSVM-3.17, and LIBSVM is used under MATLAB platforms.LIBSVM provides many default parameters, silent merely with these
The parameter recognized can be resolved various problems, but also provide cross validation function.The software can effectively solve the problem that a variety of
The problems such as classification and recurrence of SVMs, it is an easy and effective, easy to use SVM software kit.
The determination of bolting and shotcrete roadway coefficient of frictional resistance prediction model parameterses
The program in above-mentioned grid data service is called to SVM prediction bolting and shotcrete roadway coefficient of frictional resistance model
80 groups of training samples are trained, and can be supported the parameter combination of vector machine prediction bolting and shotcrete roadway coefficient of frictional resistance model
C and g, as shown in Figure 8.
As can be drawn from Figure 8, using grid data service to SVM prediction bolt-spary supports tunnel frictional resistance
When punishment parameter C and Radial basis kernel function parameter the g combination of Modulus Model are chosen, obtain its fitness value, i.e. cross validation
Mean square deviation CVMSE=0.0079583;Relative coefficient is 0.99727, and correlation is very higher;SVM prediction anchor rete cord
The best parameter group BestC=4, Bestg=2 of supported laneway coefficient of frictional resistance model.
The program in above-mentioned genetic algorithm is called to the 80 of SVM prediction bolting and shotcrete roadway coefficient of frictional resistance model
Group training sample is trained, and can be supported the parameter combination C of vector machine prediction bolting and shotcrete roadway coefficient of frictional resistance model
And g, as shown in Figure 9.
As can be drawn from Figure 9, using genetic algorithm to SVM prediction bolting and shotcrete roadway coefficient of frictional resistance mould
When punishment parameter C and Radial basis kernel function parameter the g combination of type are chosen, it is known that, with the continuous progress of iteration, fitness value exists
It is continuously getting smaller and smaller, fitness CVMSE reaches minimum value, i.e. CVMSE=0.0095713 when evolving to for 5 generation;Correlation
Coefficient is 0.99931, and correlation is very high;The best parameter group of SVM prediction bolting and shotcrete roadway coefficient of frictional resistance model
BestC=26.0592, Bestg=2.5882.
The program in above-mentioned particle cluster algorithm is called to SVM prediction bolting and shotcrete roadway coefficient of frictional resistance model
80 groups of training samples are trained, and can be supported the parameter combination of vector machine prediction bolting and shotcrete roadway coefficient of frictional resistance model
C and g, as shown in Figure 10.
As can be drawn from Figure 10, using particle cluster algorithm to SVM prediction bolting and shotcrete roadway coefficient of frictional resistance
When punishment parameter C and Radial basis kernel function parameter the g combination of model are chosen, it is known that, with the continuous progress of iteration, fitness value
Change constantly, fitness CVMSE reaches minimum value, i.e. CVMSE=0.0082523 when evolving to for 17 generation;Phase
It is 0.99716 to close property coefficient, and correlation is very high;The optimized parameter of SVM prediction bolting and shotcrete roadway coefficient of frictional resistance model
Combine BestC=15.0285, Bestg=1.1137.
The determination of roadway supported by bolt coefficient of frictional resistance prediction model parameterses
The program in above-mentioned grid data service is called to SVM prediction roadway supported by bolt coefficient of frictional resistance mould
30 groups of training samples of type are trained, and can be supported vector machine prediction roadway supported by bolt coefficient of frictional resistance model
Parameter combination C and g, as shown in figure 11.
It can be drawn from Figure 11, using grid data service to SVM prediction roadway supported by bolt frictional resistance
When punishment parameter C and Radial basis kernel function parameter the g combination of Modulus Model are chosen, obtain its fitness value, i.e. cross validation
Mean square deviation CVMSE=0.063409;Relative coefficient is 0.97443, and correlation is very higher;Roadway supported by bolt frictional resistance system
The best parameter group BestC=0.70711, Bestg=5.6569 of number forecast model.
The program in above-mentioned genetic is called to SVM prediction roadway supported by bolt coefficient of frictional resistance model
30 groups of training samples are trained, and can be supported the parameter of vector machine prediction roadway supported by bolt coefficient of frictional resistance model
C and g is combined, as shown in figure 12.
It can be drawn from Figure 12, using genetic algorithm to SVM prediction roadway supported by bolt frictional resistance system
When punishment parameter C and Radial basis kernel function parameter the g combination of exponential model are chosen, it is known that, with the continuous progress of iteration, fitness
Value is being continuously getting smaller and smaller, and fitness CVMSE reaches minimum value, i.e. CVMSE=0.062378 when evolving to for 23 generation;Phase
It is 0.97462 to close property coefficient, and correlation is very higher;SVM prediction roadway supported by bolt coefficient of frictional resistance model is most
Excellent parameter combination BestC=3.9922, Bestg=4.9293.
The program in particle swarm optimization is called to the 30 of SVM prediction roadway supported by bolt coefficient of frictional resistance model
Group training sample is trained, and can be supported the parameter group of vector machine prediction roadway supported by bolt coefficient of frictional resistance model
C and g is closed, as shown in figure 13.
It can be drawn from Figure 13, using particle cluster algorithm to SVM prediction roadway supported by bolt frictional resistance
When punishment parameter C and Radial basis kernel function parameter the g combination of Modulus Model are chosen, it is known that, with the continuous progress of iteration, adapt to
Angle value changes constantly, and fitness CVMSE reaches minimum value, i.e. CVMSE=0.062662 when evolving to for 58 generation;
Relative coefficient is 0.9742, and correlation is very high;SVM prediction roadway supported by bolt coefficient of frictional resistance model is most
Excellent parameter combination BestC=3.0257, Bestg=5.4785.
The determination of bolt-mesh-anchor support tunnel coefficient of frictional resistance prediction model parameterses
The program in above-mentioned grid data service is called to SVM prediction bolt-mesh-anchor support tunnel coefficient of frictional resistance
30 groups of training samples of model are trained, and can be supported vector machine prediction bolt-mesh-anchor support tunnel coefficient of frictional resistance mould
The parameter combination C and g of type, as shown in figure 14.
It can be drawn from Figure 14, using grid data service in the resistance that rubbed to SVM prediction bolt-mesh-anchor support tunnel
When punishment parameter C and Radial basis kernel function parameter the g combination of force coefficient model are chosen, its fitness value, i.e. cross validation are obtained
Mean square deviation CVMSE=0.05353;Relative coefficient is 0.95248, and correlation is very higher;SVM prediction anchor rete cord
The best parameter group BestC=11.3137, Bestg=0.125 of supported laneway coefficient of frictional resistance model.
The program in above-mentioned genetic algorithm is called to SVM prediction bolt-mesh-anchor support tunnel coefficient of frictional resistance mould
30 groups of training samples of type are trained, and can be supported vector machine prediction bolt-mesh-anchor support tunnel coefficient of frictional resistance model
Parameter combination C and g, as shown in figure 15.
It can be drawn from Figure 15, using genetic algorithm to SVM prediction bolt-mesh-anchor support tunnel frictional resistance
When punishment parameter C and Radial basis kernel function parameter the g combination of Modulus Model are chosen, it is known that, with the continuous progress of iteration, adapt to
Angle value is being continuously getting smaller and smaller, and fitness CVMSE reaches minimum value, i.e. CVMSE=0.56465 when evolving to for 9 generation;Phase
It is 0.95891 to close property coefficient, and correlation is very higher;SVM prediction bolt-mesh-anchor support tunnel coefficient of frictional resistance model
Best parameter group BestC=11.7466, Bestg=2.1534.
The program in above-mentioned particle cluster algorithm is called to SVM prediction bolt-mesh-anchor support tunnel coefficient of frictional resistance
30 groups of training samples of model are trained, and can be supported vector machine prediction bolt-mesh-anchor support tunnel coefficient of frictional resistance mould
The parameter combination C and g of type, as shown in figure 16.
It can be drawn from Figure 16, using particle cluster algorithm in the resistance that rubbed to SVM prediction bolt-mesh-anchor support tunnel
When punishment parameter C and Radial basis kernel function parameter the g combination of force coefficient model are chosen, it is known that, with the continuous progress of iteration, fit
Angle value is answered to change constantly, fitness CVMSE reaches minimum value, i.e. CVMSE=when evolving to for 4 generation
0.053819;Relative coefficient is 0.95507, and correlation is very high;SVM prediction bolt-mesh-anchor support tunnel frictional resistance
The best parameter group BestC=17.2549, Bestg=0.10179 of Modulus Model.
The prediction result of tunnel coefficient of frictional resistance and comparative analysis
The best parameter group BestC for the bolting and shotcrete roadway coefficient of frictional resistance forecast model chosen using trellis search method
=4, Bestg=2,50 groups of checking collection data are predicted, obtain bolting and shotcrete roadway coefficient of frictional resistance prediction result, such as Figure 17
It is shown;And contrasted with the bolting and shotcrete roadway coefficient of frictional resistance of actual measurement, predicted value and the relative error of measured value are obtained, is such as schemed
Shown in 18.
Analysis chart 17 and Figure 18's understands, the SVM prediction bolting and shotcrete roadway friction chosen using trellis search method
The best parameter group of resistance coefficient model concentrates sample to be predicted checking, shows good fitting effect.By right
The bolting and shotcrete roadway coefficient of frictional resistance value of prediction with the bolting and shotcrete roadway coefficient of frictional resistance value of actual measurement contrast, and is predicted
Relative error be respectively less than 5%, average relative error 2.703%, Averaged Square Error of Multivariate MSE=0.004848.Therefore, grid is searched
The parameter that rope method is chosen is high to the predictablity rate of bolting and shotcrete roadway coefficient of frictional resistance, and prediction effect is fine.
The best parameter group BestC for the bolting and shotcrete roadway coefficient of frictional resistance forecast model chosen using genetic algorithms approach
=26.0592, Bestg=2.5882,50 groups of checking collection data are predicted, obtain the prediction of bolting and shotcrete roadway coefficient of frictional resistance
As a result, as shown in figure 19;And contrasted with the bolting and shotcrete roadway coefficient of frictional resistance of actual measurement, obtain predicted value and the phase of measured value
To error, as shown in figure 20.
Analysis chart 19 and Figure 20's understands, the SVM prediction bolting and shotcrete roadway frictional resistance chosen using genetic algorithm
The best parameter group of Modulus Model concentrates sample to be predicted checking, shows good fitting effect.By to prediction
Bolting and shotcrete roadway coefficient of frictional resistance value with the bolting and shotcrete roadway coefficient of frictional resistance value surveyed contrast, the phase predicted
It is respectively less than 5% to error, average relative error 2.569%, Averaged Square Error of Multivariate MSE=0.004749.Therefore, genetic algorithm is selected
The parameter taken is higher to the predictablity rate of bolting and shotcrete roadway coefficient of frictional resistance, and prediction effect is fine.
The best parameter group BestC=15.0285, Bestg=for the forecast model chosen using particle cluster algorithm method
1.1137,50 groups of checking collection data are predicted, obtain bolting and shotcrete roadway coefficient of frictional resistance prediction result, as shown in figure 21;
And contrasted with the bolting and shotcrete roadway coefficient of frictional resistance of actual measurement, predicted value and the relative error of measured value are obtained, such as Figure 22 institutes
Show.
Analysis chart 21 and Figure 22's understands that the SVM prediction bolting and shotcrete roadway chosen using particle cluster algorithm method rubs
The best parameter group for wiping resistance coefficient model concentrates sample to be predicted checking, shows good fitting effect.Pass through
To the bolting and shotcrete roadway coefficient of frictional resistance value of prediction and contrasting for the bolting and shotcrete roadway coefficient of frictional resistance value of actual measurement, obtain pre-
The relative error of survey is respectively less than 5%, average relative error 2.746%, Averaged Square Error of Multivariate MSE=0.0049279.Therefore, grain
The parameter that swarm optimization is chosen is high to the predictablity rate of bolting and shotcrete roadway coefficient of frictional resistance, and prediction effect is fine.
The analysis of bolting and shotcrete roadway coefficient of frictional resistance prediction result
Using grid data service, genetic algorithm, particle cluster algorithm to SVM prediction bolting and shotcrete roadway frictional resistance system
Exponential model parameter combination --- punishment parameter C and Radial basis kernel function parameter g chooses result such as table 4.
The parameter combination performance comparison that 4 three kinds of methods of table are chosen
Coefficient correlation is can be seen that from the result of prediction and is attained by more than 99.7%, is calculated based on grid data service, heredity
The SVM prediction bolting and shotcrete roadway coefficient of frictional resistance model of these three method Selecting All Parameters of method and particle cluster algorithm, can
Initial data is fitted well;Mean square error is respectively less than 0.01, and wherein the performance of genetic algorithms approach regression fit is most
It is good;Relative error is respectively less than 5%, and average relative error is respectively less than 3%, the bolting and shotcrete roadway friction chosen based on genetic algorithms approach
The parameter combination of resistance coefficient forecast model, the average relative error of prediction is minimum, is 2.569%;Choose bolting and shotcrete roadway friction
The optimal adaptation degree CVMSE of resistance coefficient prediction model parameterses combination is also that genetic algorithms approach behaves oneself best.By to shotcrete
The result that tunnel coefficient of frictional resistance is predicted shows that the SVMs based on the combination of genetic algorithms approach Selecting All Parameters is pre-
Bolting and shotcrete roadway coefficient of frictional resistance model performance is surveyed to behave oneself best.Therefore, SVM prediction anchor genetic algorithm chosen
The parameter BestC=26.0592, Bestg=2.5882 for spraying tunnel coefficient of frictional resistance model are applied to bolting and shotcrete roadway friction resistance
In force coefficient, good prediction result can be obtained.
Roadway supported by bolt coefficient of frictional resistance prediction result and comparative analysis
The best parameter group for the roadway supported by bolt coefficient of frictional resistance forecast model chosen using trellis search method
BestC=0.70711, Bestg=5.6569,15 groups of checking collection data are predicted, obtain roadway supported by bolt friction resistance
Force coefficient prediction result, as shown in figure 23;And contrasted with the roadway supported by bolt coefficient of frictional resistance of actual measurement, predicted
The relative error of value and actual value, as shown in figure 24.
Analysis chart 23 and Figure 24's understands, the SVM prediction suspension roof support lane chosen using grid data service method
The best parameter group of road coefficient of frictional resistance model concentrates sample to be predicted checking, shows good fitting effect.
Entered by roadway supported by bolt coefficient of frictional resistance value to prediction with the roadway supported by bolt coefficient of frictional resistance value surveyed
Row contrast, the relative error predicted are respectively less than 7%, average relative error 4.84%, Averaged Square Error of Multivariate MSE=
0.0091416.Therefore, the mould accuracy rate predicted roadway supported by bolt coefficient of frictional resistance of parameter that grid data service is chosen compared with
Height, prediction effect are fine.
The best parameter group for the roadway supported by bolt coefficient of frictional resistance forecast model chosen using genetic algorithms approach
BestC=3.9922, Bestg=4.9293,15 groups of checking collection data are predicted, obtain roadway supported by bolt frictional resistance
Coefficient prediction result, as shown in figure 25;And contrasted with the roadway supported by bolt coefficient of frictional resistance of actual measurement, obtain predicted value
With the relative error of actual measured value, as shown in figure 26.
Analysis chart 25 and Figure 26's understands, the SVM prediction roadway supported by bolt chosen using genetic algorithms approach
The best parameter group of coefficient of frictional resistance model concentrates sample to be predicted checking, shows good fitting effect.It is logical
Cross the progress to the roadway supported by bolt coefficient of frictional resistance value and the roadway supported by bolt coefficient of frictional resistance value of actual measurement of prediction
Contrast, the relative error predicted are respectively less than 6%, average relative error 4.54%, Averaged Square Error of Multivariate MSE=
0.0091382.Therefore, the accuracy rate that the parameter that genetic algorithm is chosen is predicted roadway supported by bolt coefficient of frictional resistance is higher,
Prediction effect is fine.
The optimized parameter group for the roadway supported by bolt coefficient of frictional resistance forecast model chosen using particle cluster algorithm method
BestC=3.0257, Bestg=5.4785 are closed, 15 groups of checking collection data are predicted, obtains roadway supported by bolt friction resistance
Force coefficient prediction result, as shown in figure 27;And contrasted with the roadway supported by bolt coefficient of frictional resistance of actual measurement, predicted
Value and the relative error of measured value, as shown in figure 28.
Analysis chart 27 and Figure 28's understands, the SVM prediction suspension roof support lane chosen using particle cluster algorithm method
The best parameter group of road coefficient of frictional resistance model concentrates sample to be predicted checking, shows good fitting effect.
Entered by roadway supported by bolt coefficient of frictional resistance value to prediction with the roadway supported by bolt coefficient of frictional resistance value surveyed
Row contrast, the relative error predicted are respectively less than 7%, average relative error 4.78%, Averaged Square Error of Multivariate MSE=
0.0091397.Therefore, particle cluster algorithm choose parameter to the accuracy rate that roadway supported by bolt coefficient of frictional resistance is predicted compared with
Height, prediction effect are fine.
The analysis of roadway supported by bolt coefficient of frictional resistance prediction result
SVM prediction roadway supported by bolt is rubbed using grid data service, genetic algorithm, particle cluster algorithm and hindered
Force coefficient model parameter combines --- and punishment parameter C and Radial basis kernel function parameter g chooses result such as table 5.
The punishment parameter and kernel functional parameter table that 5 three kinds of methods of table are chosen
Coefficient correlation is can be seen that from the result of prediction and is attained by more than 97.4%, is calculated based on grid data service, heredity
The SVM prediction roadway supported by bolt coefficient of frictional resistance model of these three method Selecting All Parameters of method and particle cluster algorithm,
Initial data can be fitted well.Mean square error is respectively less than 0.01, the wherein table of genetic algorithms approach regression fit
It is now best;Relative error is respectively less than 7%, and average relative error is respectively less than 5%, the suspension roof support lane that genetic algorithms approach is chosen
The parameter combination of road coefficient of frictional resistance forecast model, the average relative error of prediction is minimum, is 4.54%;Choose suspension roof support
The optimal adaptation degree CVMSE of tunnel coefficient of frictional resistance prediction model parameterses combination is also that genetic algorithms approach behaves oneself best.It is logical
Cross the result for being predicted roadway supported by bolt coefficient of frictional resistance to show, based on the combination of genetic algorithms approach Selecting All Parameters
SVM prediction roadway supported by bolt coefficient of frictional resistance model performance behaves oneself best.Therefore, genetic algorithm is chosen
The parameter BestC=3.9922, Bestg=4.9293 of SVM prediction roadway supported by bolt coefficient of frictional resistance model should
For in roadway supported by bolt coefficient of frictional resistance, good prediction result can be obtained.
Bolt-mesh-anchor support tunnel coefficient of frictional resistance prediction result and comparative analysis
The best parameter group BestC=11.3137, Bestg=for the forecast model chosen using trellis search method
0.125,20 groups of checking collection data are predicted, obtain anchor rete cord tunnel coefficient of frictional resistance prediction result, as shown in figure 29;
And contrasted with the bolt-mesh-anchor support tunnel coefficient of frictional resistance of actual measurement, predicted value and the relative error of measured value are obtained, such as
Shown in Figure 30.
Analysis chart 29 and Figure 30's understands, the SVM prediction bolt-mesh-anchor support lane chosen using trellis search method
The best parameter group of road coefficient of frictional resistance model concentrates sample to be predicted checking, shows good fitting effect.
Pass through the bolt-mesh-anchor support tunnel coefficient of frictional resistance value to prediction and the bolt-mesh-anchor support tunnel coefficient of frictional resistance value of actual measurement
Carry out contrast, the relative error predicted is respectively less than 6%, average relative error 4.92%, Averaged Square Error of Multivariate MSE=
0.008925.Therefore, grid data service choose parameter to the accuracy rate that bolt-mesh-anchor support tunnel coefficient of frictional resistance is predicted compared with
Height, prediction effect are fine.
The optimized parameter group for the bolt-mesh-anchor support tunnel coefficient of frictional resistance forecast model chosen using genetic algorithms approach
BestC=11.7466, Bestg=2.1534 are closed, 20 groups of checking collection data are predicted, bolt-mesh-anchor support tunnel is obtained and rubs
Resistance coefficient prediction result is wiped, as shown in figure 31;And contrasted with the bolt-mesh-anchor support tunnel coefficient of frictional resistance of actual measurement, obtain
To predicted value and the relative error of measured value, as shown in figure 32.
Analysis chart 31 and Figure 32's understands, the SVM prediction bolt-mesh-anchor support lane chosen using genetic algorithms approach
The best parameter group of road coefficient of frictional resistance model concentrates sample to be predicted checking, shows good fitting effect.
Pass through the bolt-mesh-anchor support tunnel coefficient of frictional resistance value to prediction and the bolt-mesh-anchor support tunnel coefficient of frictional resistance value of actual measurement
Carry out contrast, the relative error predicted is respectively less than 6%, average relative error 4.22%, Averaged Square Error of Multivariate MSE=
0.008107.Therefore, the accuracy rate that the parameter that genetic algorithm is chosen is predicted bolt-mesh-anchor support tunnel coefficient of frictional resistance is higher,
Prediction effect is fine.
The optimized parameter for the bolt-mesh-anchor support tunnel coefficient of frictional resistance forecast model chosen using particle cluster algorithm method
BestC=17.2549, Bestg=0.10179 are combined, 20 groups of checking collection data are predicted, obtain bolt-mesh-anchor support tunnel
Coefficient of frictional resistance prediction result, as shown in figure 33;And contrasted with the bolt-mesh-anchor support tunnel coefficient of frictional resistance of actual measurement,
Predicted value and the relative error of measured value are obtained, as shown in figure 34.
Analysis chart 33 and Figure 34's understands, the SVM prediction bolt-mesh-anchor support chosen using particle cluster algorithm method
The best parameter group of tunnel coefficient of frictional resistance model concentrates sample to be predicted checking, shows fitting effect well
Fruit.Pass through the bolt-mesh-anchor support tunnel coefficient of frictional resistance value to prediction and the bolt-mesh-anchor support tunnel coefficient of frictional resistance of actual measurement
Value contrast, and the relative error predicted is respectively less than 6%, average relative error 4.55%, Averaged Square Error of Multivariate MSE=
0.009692.Therefore, the parameter that particle cluster algorithm is chosen is higher to the accuracy rate of bolt-mesh-anchor support tunnel coefficient of frictional resistance, in advance
It is fine to survey effect.
The analysis of bolt-mesh-anchor support tunnel coefficient of frictional resistance prediction result
SVM prediction bolt-mesh-anchor support tunnel is rubbed using grid data service, genetic algorithm, particle cluster algorithm
Resistance coefficient model parameter combines --- and punishment parameter C and Radial basis kernel function parameter g chooses result such as table 6.
The parameter combination performance comparison that 6 three kinds of methods of table are chosen
Coefficient correlation is can be seen that from the result of prediction and is attained by more than 95.2%, is calculated based on grid data service, heredity
The SVM prediction bolt-mesh-anchor support tunnel coefficient of frictional resistance mould of these three method Selecting All Parameters of method and particle cluster algorithm
Type, initial data can be fitted well.Mean square error is respectively less than 0.01, wherein genetic algorithms approach regression fit
Behave oneself best;Relative error is respectively less than 6%, and average relative error is respectively less than 5%, the prediction anchor net that genetic algorithms approach is chosen
The parameter combination of rope supported laneway coefficient of frictional resistance model, the average relative error of prediction is minimum, is 4.22%;Choose anchor net
The optimal adaptation degree CVMSE of rope supported laneway coefficient of frictional resistance prediction model parameterses combination is also that genetic algorithms approach shows most
It is good.Therefore, the parameter for SVM prediction bolt-mesh-anchor support tunnel coefficient of frictional resistance model genetic algorithm chosen
BestC=11.7466, Bestg=2.1534 are applied in the coefficient of frictional resistance of bolt-mesh-anchor support tunnel, can obtain well
Prediction result.
Comprehensive SVMs is to bolting and shotcrete roadway coefficient of frictional resistance prediction result, roadway supported by bolt coefficient of frictional resistance
Knowable to prediction result and bolt-mesh-anchor support tunnel coefficient of frictional resistance prediction result, the model based on genetic algorithms approach selection
Parameter combination behaves oneself best to SVM prediction mine laneway coefficient of frictional resistance model performance.
This specific implementation with cold water well colliery bolting and shotcrete roadway, roadway supported by bolt, bolt-mesh-anchor support tunnel frictional resistance
Coefficient is research object, and training set is instructed with trellis search method, genetic algorithms approach, particle cluster algorithm method respectively
Practice, obtain optimal Radial basis kernel function parameter and punishment parameter combination, forecast model is established using the parameter combination of selection, it is right
Checking collection is predicted, and is obtained to draw a conclusion:
(1) parameter combination chosen using trellis search method, genetic algorithms approach, particle cluster algorithm method is established pre-
The relative error for surveying the mine laneway coefficient of frictional resistance of model prediction is respectively less than 7%, and prediction effect is preferable.
(2) the SVM prediction mine laneway coefficient of frictional resistance model prediction knot of genetic algorithm Selecting All Parameters combination
The prediction result that fruit chooses the forecast model of Selecting All Parameters combination compared with grid data service, particle cluster algorithm is compared, average to miss relatively
Difference and mean square error are minimum, fitness is optimal, degree of fitting highest, and combination property behaves oneself best.
Described above is only the preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, under the premise without departing from the principles of the invention, some improvements and modifications can also be made, these improvements and modifications also should
It is considered as protection scope of the present invention.
Claims (2)
1. a kind of mine laneway coefficient of frictional resistance Forecasting Methodology based on SVMs, it is characterised in that including following step
Suddenly:
S1, the input/output variable for determining SVM prediction mine laneway coefficient of frictional resistance, real data is gathered, and it is right
Real data is normalized, and obtains the training sample data for forecast model;
S2, choose kernel function of the Radial basis kernel function as SVM prediction mine laneway coefficient of frictional resistance model;
S3, using genetic algorithm the hyper parameter of forecast model is determined, and whole training set is trained, supported
Vector machine predicts the model of mine laneway coefficient of frictional resistance;
S4, the data concentrated using the forecast model that step S3 is established to checking are predicted.
2. a kind of mine laneway coefficient of frictional resistance Forecasting Methodology based on SVMs as claimed in claim 1, it is special
Sign is that the genetic algorithm is chosen SVM prediction model parameter combination and completed by following steps:
S31, the coding of chromosome, decoding and the initialization of population
Using the coded system based on real number, punishment parameter C and kernel functional parameter the g search of SVM prediction model are set
Scope, then the parameter combination value with feasibility in setting range is converted to the dyeing that can be received by genetic algorithm
Body, n are finally randomly selected in initial data and carries out initialization process with to completing population;
S32, determine population at individual fitness
Using the mean square error of forecast model in the training process as individual adaptation degree evaluation index, i.e., some chromosome in population
Relative to the superiority-inferiority of other chromosomes during evolution;
S33, the selection of chromosome, intersection and variation
Individual is selected according to championship principle, after the select probability for giving individual, can produce between [0,1] it is uniform with
Machine number, these random numbers can determine the mating probability of individual, the individual that vitality is strong in population, fitness is high is participated in new colony
Generation, while the individual in original colony is selected the superior and eliminated the inferior, approach the equal iteration of fitness value of each individual in colony
Globally optimal solution.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711032948.5A CN107729671A (en) | 2017-10-25 | 2017-10-25 | A kind of mine laneway coefficient of frictional resistance Forecasting Methodology based on SVMs |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711032948.5A CN107729671A (en) | 2017-10-25 | 2017-10-25 | A kind of mine laneway coefficient of frictional resistance Forecasting Methodology based on SVMs |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107729671A true CN107729671A (en) | 2018-02-23 |
Family
ID=61203215
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711032948.5A Pending CN107729671A (en) | 2017-10-25 | 2017-10-25 | A kind of mine laneway coefficient of frictional resistance Forecasting Methodology based on SVMs |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107729671A (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108920310A (en) * | 2018-05-23 | 2018-11-30 | 携程旅游网络技术(上海)有限公司 | The rejecting outliers method and system of interface data |
CN109753624A (en) * | 2019-01-10 | 2019-05-14 | 天地(常州)自动化股份有限公司 | Mensuration of Mine Ventilation Resistance method based on feature tunnel |
CN109854299A (en) * | 2018-12-11 | 2019-06-07 | 中煤科工集团重庆研究院有限公司 | Method for rapidly determining friction resistance coefficient of ventilation roadway based on big data |
CN110321588A (en) * | 2019-05-10 | 2019-10-11 | 中车青岛四方车辆研究所有限公司 | Rail vehicle aerodynamic Drag Calculation method based on numerical simulation |
CN110717528A (en) * | 2019-09-25 | 2020-01-21 | 中国石油大学(华东) | Support vector machine-based sedimentary microfacies identification method using conventional logging information |
CN111260077A (en) * | 2020-01-14 | 2020-06-09 | 支付宝(杭州)信息技术有限公司 | Method and device for determining hyper-parameters of business processing model |
CN111444629A (en) * | 2020-04-15 | 2020-07-24 | 中国二冶集团有限公司 | Reinforcing steel bar corrosion parameter prediction method based on support vector machine |
CN112765891A (en) * | 2021-01-27 | 2021-05-07 | 辽宁工程技术大学 | Method for predicting maximum value of mine fire disaster factor |
CN113627080A (en) * | 2021-07-29 | 2021-11-09 | 西安科技大学 | Coal mine driving face air quantity demand prediction method based on support vector machine |
CN116401931A (en) * | 2023-06-08 | 2023-07-07 | 吉林大学 | Circulation well structure and operation parameter optimization method, system and equipment |
CN116702096A (en) * | 2023-08-04 | 2023-09-05 | 中汽研汽车检验中心(昆明)有限公司 | Method and device for measuring and calculating road sliding resistance of vehicle plateau environment |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102903007A (en) * | 2012-09-20 | 2013-01-30 | 西安科技大学 | Method for optimizing disaggregated model by adopting genetic algorithm |
US8688603B1 (en) * | 2011-11-14 | 2014-04-01 | Amazon Technologies, Inc. | System and method for identifying and correcting marginal false positives in machine learning models |
-
2017
- 2017-10-25 CN CN201711032948.5A patent/CN107729671A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8688603B1 (en) * | 2011-11-14 | 2014-04-01 | Amazon Technologies, Inc. | System and method for identifying and correcting marginal false positives in machine learning models |
CN102903007A (en) * | 2012-09-20 | 2013-01-30 | 西安科技大学 | Method for optimizing disaggregated model by adopting genetic algorithm |
Non-Patent Citations (1)
Title |
---|
董晓雷: "基于GA-SVM的煤层瓦斯涌出量预测技术研究", 《中国优秀硕士学位论文全文数据库工程科技Ⅰ辑(月刊 )》 * |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108920310B (en) * | 2018-05-23 | 2022-05-03 | 携程旅游网络技术(上海)有限公司 | Abnormal value detection method and system of interface data |
CN108920310A (en) * | 2018-05-23 | 2018-11-30 | 携程旅游网络技术(上海)有限公司 | The rejecting outliers method and system of interface data |
CN109854299A (en) * | 2018-12-11 | 2019-06-07 | 中煤科工集团重庆研究院有限公司 | Method for rapidly determining friction resistance coefficient of ventilation roadway based on big data |
CN109854299B (en) * | 2018-12-11 | 2020-10-23 | 中煤科工集团重庆研究院有限公司 | Method for rapidly determining friction resistance coefficient of ventilation roadway based on big data |
CN109753624A (en) * | 2019-01-10 | 2019-05-14 | 天地(常州)自动化股份有限公司 | Mensuration of Mine Ventilation Resistance method based on feature tunnel |
CN109753624B (en) * | 2019-01-10 | 2023-03-24 | 天地(常州)自动化股份有限公司 | Mine ventilation resistance measuring method based on characteristic roadway |
CN110321588A (en) * | 2019-05-10 | 2019-10-11 | 中车青岛四方车辆研究所有限公司 | Rail vehicle aerodynamic Drag Calculation method based on numerical simulation |
CN110717528A (en) * | 2019-09-25 | 2020-01-21 | 中国石油大学(华东) | Support vector machine-based sedimentary microfacies identification method using conventional logging information |
CN111260077A (en) * | 2020-01-14 | 2020-06-09 | 支付宝(杭州)信息技术有限公司 | Method and device for determining hyper-parameters of business processing model |
CN111444629A (en) * | 2020-04-15 | 2020-07-24 | 中国二冶集团有限公司 | Reinforcing steel bar corrosion parameter prediction method based on support vector machine |
CN112765891A (en) * | 2021-01-27 | 2021-05-07 | 辽宁工程技术大学 | Method for predicting maximum value of mine fire disaster factor |
CN112765891B (en) * | 2021-01-27 | 2023-11-10 | 辽宁工程技术大学 | Prediction method for maximum value of disaster-causing factors of mine fire disaster |
CN113627080A (en) * | 2021-07-29 | 2021-11-09 | 西安科技大学 | Coal mine driving face air quantity demand prediction method based on support vector machine |
CN113627080B (en) * | 2021-07-29 | 2023-06-20 | 西安科技大学 | Coal mine tunneling working face air quantity demand prediction method based on support vector machine |
CN116401931A (en) * | 2023-06-08 | 2023-07-07 | 吉林大学 | Circulation well structure and operation parameter optimization method, system and equipment |
CN116401931B (en) * | 2023-06-08 | 2023-08-18 | 吉林大学 | Circulation well structure and operation parameter optimization method, system and equipment |
CN116702096A (en) * | 2023-08-04 | 2023-09-05 | 中汽研汽车检验中心(昆明)有限公司 | Method and device for measuring and calculating road sliding resistance of vehicle plateau environment |
CN116702096B (en) * | 2023-08-04 | 2023-10-03 | 中汽研汽车检验中心(昆明)有限公司 | Method and device for measuring and calculating road sliding resistance of vehicle plateau environment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107729671A (en) | A kind of mine laneway coefficient of frictional resistance Forecasting Methodology based on SVMs | |
Liu et al. | A future land use simulation model (FLUS) for simulating multiple land use scenarios by coupling human and natural effects | |
Bai et al. | Simulating runoff under changing climatic conditions: a comparison of the long short-term memory network with two conceptual hydrologic models | |
Li et al. | Exploration of China's net CO2 emissions evolutionary pathways by 2060 in the context of carbon neutrality | |
CN109461025B (en) | Electric energy substitution potential customer prediction method based on machine learning | |
CN112785450B (en) | Soil environment quality partitioning method and system | |
CN110225055A (en) | A kind of network flow abnormal detecting method and system based on KNN semi-supervised learning model | |
Fu et al. | An innovative decision making method for air quality monitoring based on big data-assisted artificial intelligence technique | |
Yu et al. | Fireworks algorithm with covariance mutation | |
CN104809476A (en) | Multi-target evolutionary fuzzy rule classification method based on decomposition | |
Eirgash et al. | A novel oppositional teaching learning strategy based on the golden ratio to solve the Time-Cost-Environmental impact Trade-off optimization problems | |
Chen et al. | Global oceanic eddy identification: A deep learning method from argo profiles and altimetry data | |
CN115033591B (en) | Intelligent detection method, system, storage medium and computer equipment for electric charge data abnormality | |
Xue et al. | PREDICTION OF SLOPE STABILITY BASED ON GA-BP HYBRID ALGORITHM. | |
CN115019166A (en) | Marsh wetland information extraction method, device, medium and terminal based on deep network model | |
Chen et al. | Hyperspectral detection of sugar content for sugar-sweetened apples based on sample grouping and SPA feature selecting methods | |
Hosseini et al. | A causality-weighted approach for prioritizing mining 4.0 strategies integrating reliability-based fuzzy cognitive map and hybrid decision-making methods: A case study of Nigerian Mining Sector | |
Wen et al. | An improved LSTM-based model for identifying high working intensity load segments of the tractor load spectrum | |
Xu et al. | Research on height prediction of water-conducting fracture zone in coal mining based on intelligent algorithm combined with extreme boosting machine | |
Niu et al. | A novel framework combining production evaluation and quantification of development parameters for shale gas wells | |
Daniel et al. | Bayesian optimization-enhanced ensemble learning for the uniaxial compressive strength prediction of natural rock and its application | |
CN114841064A (en) | Drought disaster weather prediction method based on semi-supervised integrated learning | |
Qu et al. | Mining Engineering Image Recognition Method Based on Simulated Annealing Algorithm | |
Ujjainia et al. | Crop Yield Prediction using Regression Model | |
Wu et al. | Multisource information fusion for real-time prediction and multiobjective optimization of large-diameter slurry shield attitude |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180223 |