CN105303251A - Interval support vector machine model for solar seawater desalination problem, and optimized design thereof - Google Patents

Abstract

The invention relates to uncertain regression and optimized design technology for a solar seawater desalination problem. Firstly, interval processing is performed on original data output so that an output result has an interval property, thereby settling uncertainty of the output result in actual engineering. Then a support vector machine is adopted for discrete data with the interval property for realizing regression, and an input interval support vector machine model and an output interval support vector machine model are established for settling high difficulty in traditional mathematical modeling. Finally, based on an interval dominance relationship, optimal solutions of input variables such as hot air temperature are calculated by means of an IP-MOEA algorithm, thereby obtaining an optimal operation condition of a solar desalination system. For facilitating input data modification and displaying an output result visually, the invention provides an uncertain regression and optimization platform for the solar seawater desalination problem based on Matlab GUI. The interval support vector machine model mainly realizes functions of displaying a parameter of parameter optimization in a regression process, displaying results of images of regression curve, outputting the EXCEL of the optimal solution set, setting parameters related with evolution generation in an optimization process, and designing buttons of file opening button, etc.

Description

The interval supporting vector machine model of solar seawater desalination problem and optimal design

Technical field

This patent belongs to field of seawater desalination, be specifically related to the content that a kind of two target support vector machine regressions about sea-water-desalination water producing rate and cost and interval multi-target evolution are optimized, can be used for solving there is bounded-but-unknown uncertainty data regression in interval multi-objective optimization question.

Background technology

Along with growth and the economic development of population, there is shortage in fresh water and conventional energy resources, for improving solar seawater desalination rate, reducing the research of fresh water cost, more and more causes the concern of Chinese scholars.For the optimization of solar seawater desalination problem, first to set up fresh water producing water ratio, produce the regression model of the input variable such as water cost and hot air temperature.At present, existing recurrence mode has polynomial regression, radial basis function, neural network, and support vector machine etc.In paper " performance simulation of hollow fiber air gap membrane distillation desalting process and optimization ", utilize Response Surface Method, mock standard seawater is to the factor of influence of hollow fiber air gap membrane distillation desalting process, and membrane flux index is optimized, designed by the optimum experimental achieved towards central complex design method based on hot feed liquid inflow temperature, condensed fluid inflow temperature and feed liquid flow, and establish the quadratic polynomial regression model between response and factor of influence; In paper " Modelingandgeneticalgorithm-basedmulti-objectiveoptimiza tionoftheMED-TVCdesalinationsystem ", for multi-effect distilling thermal vapor compression desalination system, adopt response surface and least square method, build annual total expenses, gain input ratio, and the agent model of fresh-water flow, and apply genetic algorithm, optimize above-mentioned three targets.In paper " Variousapproachesinoptimizationofmultieffectsdistillatio ndesalinationsystemsusingahybridmeta-heuristicoptimizati ontool ", for the multi-stage distilled seawater desalination system based on vapor compressor, adopt energy, entropy analysis and gross income respectively, set up thermodynamics and the Economic Model of this system, and utilize the hybrid algorithm formed by genetic algorithm and simulated annealing, optimize above-mentioned model.

The effective way optimizing seawater desalination system improves the utilization ratio of the energy, reduction desalinating cost, for targets different in desalinization, also has different Optimized models.In paper " optimal design of reverse osmosis seawater desalination system ", the design of reverse osmosis seawater desalting process is studied.First for each unit of process, unit operations model and relevant economic model is given; Then, by certain variable by interrelated for these models, and system model is formed; Finally, minimum for objective function, process thermodynamics, lectotype selection, designing requirement etc. are for constraint with annual cost, optimize reverse osmosis seawater desalination system; In paper " novel solar seawater desalination device principle of work and performance optimization ", in order to improve the performance of device, the power consumption consumed with unit fresh water yield is minimum is objective function, establish the mathematical model that device performance parameters is optimized, obtain the Changing Pattern of optimum evaporating temperature device power consumption with evaporating temperature.In paper " Multi-objectiveoptimizationofamembranedistillationsystem fordesalinationofseawater ", Sharma and Rangaiah, by maximizing producing water ratio and minimizing energy resource consumption, optimizes film distilling seawater desalination system.In paper " Multi-objectivedesignofreverseosmosisplantsintergratedwi thsolarrankinecyclesandthermalenergystorage ", by considering cost under given water requirements condition and environmental impact, set up the multiple goal mixed-integer nonlinear programming model of integrated system, and adopt ε leash law, solve above-mentioned model.

Simultaneously, in patent CN102163249A, propose and utilize interactive evolution optimization method for solving the GUI Platform Designing of curtain design, in interactive process, for user provides multiple individual evaluation, exact value evaluation and automatic Evaluation, during evolution, be supported between Different Evolutionary generation and use different coded systems and corresponding evolution algorithm, facilitate user to search for curtain scheme in different regions.

Above-mentioned technological achievement is the foundation of desalinization problem model, the optimization of operating condition provides practicable method, but it should be noted that, still there is following deficiency in existing method: (1) is current, the main variable measured of desalinization experiment has, the temperature of hot-air, environment temperature, the flow of dry air, evaporator outlet temperature and water capacity, condensator outlet temperature and water capacity, because internal math relation is complicated, and the fresh water production etc. exported has certain uncertainty, for the relation between this multi input variable and uncertain output variable, so far the uncertain regression model setting up desalinization problem is seldom had, (2) existing method is all set up desalinization problem order (many) objective optimization model really, and employing determines that optimization method solves, in view of this problem has certain uncertainty, and there is conflicting output target, be necessary to build its uncertain Model for Multi-Objective Optimization, and adopt uncertain evolution optimization method to solve, (3) in the uncertain recurrence of desalinization and optimizing process, there is the setting of some parameters and the output of result, for the personnel being unfamiliar with program code, be difficult to modify to it.Therefore, set up visual GUI platform, be absolutely necessary.

Summary of the invention

The object of the invention is for the problems referred to above, design one about the uncertain recurrence of solar seawater desalination and the GUI platform of optimization, set up the input variables such as hot air temperature and the Interval Regression model between desalinization rate and cost two targets, and the input variables such as hot air temperature are optimized.

Technical matters to be solved by this invention comprises: how original output data are carried out interval by (1), set up Interval Regression model; (2) how to optimize conflicting two uncertain targets, to obtain optimized operation operating mode collection simultaneously; (3) how easily parameters and display translation result.

The solution of the technology of the present invention is: first, makes full use of primary data information (pdi), using having the maximum output valve of identical input data as the output upper bound, using minimum output valve as lower bound, thus realizes the interval of data; Secondly, at the built-in n-fold crosscheck in LibSVM tool box model, by arranging the parameters such as crosscheck step-length-v, optimized parameter can be obtained by the mode of optimizing; Then, set up interval two objective optimization model based on producing water ratio and cost, to produce the input data such as hot air temperature at random for initial population, utilize the Interval Regression model built, obtain that there are two regional computer targets to export, and adopt the IP-MOEA algorithm be dominant based on interval, find the optimal solution set of input variable; Finally, build the GUI platform of desalinization experiment, parameters is made input dialogue frame, result is shown as output pattern interface.It is characterized in that:

(1) the solar seawater desalination Interval Regression model based on LibSVM tool box is set up

Solar seawater desalination experiment relate generally to following parameter: the temperature of hot-air, the temperature of chilled water, hot-air flow velocity, dry air, evaporator outlet temperature, evaporator outlet water capacity, condensator outlet temperature, condensator outlet water capacity.Because internal relations is complicated, is not easy to utilize existing mathematical model to go to build its mapping relations, utilizes LibSVM tool box can set up the Interval Regression model of desalination rate and its dependent variable easily and efficiently.

Build solar seawater desalination Interval Regression model and have following step:

Step1: tested by desalinization, obtains raw data, composing training collection data and test set data;

Step2: training set is exported data interval, obtains the interval upper bound and lower bound respectively;

Step3: training set data is normalized;

Step4: the circulation step-length that the penalty-c in LibSVM tool box, gamma function-g, kernel function type-t, n-fold cross-verification model-v and SVR types of models-s five are set;

Step5: the error calculating crosscheck by svmtrain, is stored in best_c, best_g, best_t, best_s respectively by-the c ,-g in least error situation ,-t ,-s parameter;

Step6: by best_c, best_g, best_t, best_s, has regional computer training set data and is set in svmtrain, and training obtains optimum upper bound model model_up, optimum lower bound model model_down;

Step7: by training set data respectively with optimum upper bound model model_up, optimum lower bound model model_down is set in svmpredict, the upper bound of respectively output regression prediction and lower bound;

Step8: in like manner, by test set data respectively with optimum upper bound model model_up, optimum lower bound model model_down is set in svmpredict, the upper bound of respectively output regression prediction and lower bound.

(2) based on the desalinization optimal design of IP-MOEA

For solar seawater desalination problem, maintain dry air flow, hot-air flow velocity, and when environment hot air temperature is constant, the fresh water productive rate of device depends primarily on soft air by the water capacity difference before and after condenser; Produce water cost to produce primarily of the power of condenser, therefore, adopt steam to be shown as this by the thermometer before and after condenser, more intuitively and be easy to measurement.In solar seawater desalination problem, due to the impact of the uncertain factors such as environment, make fresh water productive rate and produce water cost that there is certain uncertainty, thus, set up interval two objective optimization model of solar seawater desalination problem, and adopt the evolution optimization method IP-MOEA algorithm of the interval multi-objective optimization question of a kind of effective solution to solve.

Optimization problem is such as formula shown in (1):

minf(x)＝(1/f ₁(x),f ₂(x))，(1)

Wherein x=(x ₁, x ₂, x ₃, x ₄) ∈ X, x ₁for the water capacity difference before and after condenser, x ₂for the flow of dry air, x ₃for the flow velocity of hot-air, x ₄for the temperature of hot-air, f ₁x () is desalinization productive rate, f ₂x () is desalinization cost, shown in (2):

1/f ₁(x)＝[1/svmpredict(model ₁_up),1/svmpredict(model ₁_down)],

f ₂(x)＝[svmpredict(model ₂_down),svmpredict(model ₂_up)],

(2)

Wherein, svmpredict () is Support vector regression function, model _i_ down and model _i_ up is respectively the lower bound and upper bound optimum regression model of training i-th target obtained.

(3) the desalinization Interval Regression based on MatlabGUI and optimal design platform is built

Based on the desalinization Interval Regression of gui interface and the major function of Optimization Platform be, realize man-machine interaction, can optimizing process parameter be set on platform, trigger experiment process, optimizing parameter in display Interval Regression, export training set, the fresh water productive rate of test set, the regression curve and the raw data comparison diagram that produce water cost, optimize after productive rate-cost forward position figure, the Microsoft Excel of output optimal solution set and its desired value.

Technique scheme has following innovative point:

(1) the interval design exported.Due in desalting process, Output rusults exists uncertain, therefore it is more objective, reliable output to be expressed as interval value.When exporting between setting area, consider the information of raw data, using the upper bound that the maximum output valve of a large amount of identical input data exports as interval, the lower bound that minimum output valve exports as interval.For the situation that there are not identical input data, original output data are added Gauss disturbance is as the output upper bound, deduct Gauss disturbance as output lower bound.By said method, more objective interval number can be obtained and export.

(2) foundation of Interval Regression model.Can inputoutput data be obtained by desalinization experiment, but not have definite mathematical model to go to describe the relation of constrained input.The present invention is based on LibSVM tool box, by establishing a kind of polytypic network, data will be exported as tag along sort, input Data classification attribute, thus set up interval supporting vector machine model, establish the contact between constrained input data.

(3) structure of interval Model for Multi-Objective Optimization.In desalting process, except considering the desalination rate of seawater, also to consider the Cost Problems of desalinization.Because when desalinization rate increases, desalinization cost also increases thereupon, so while consideration productive rate, the restriction of cost also must be considered.Therefore, there are two objective optimisation problems about productive rate and cost in desalting process.The present invention utilizes above-mentioned two objective optimisation problems of IP-MOEA Algorithm for Solving be dominant based on interval, finds the optimal solution set of operating condition.

(4) design of visible human machine interactive interface.The present invention returns at uncertain data and in interval multiple-objection optimization process, relates to a large amount of optimum configurations and Output rusults, by design interactive interface, can facilitate parameters intuitively, show inner optimizing result, and output regression and optimum results.

Accompanying drawing explanation

Fig. 1 is the interval supporting vector machine model of solar seawater desalination problem and the process flow diagram of optimal design;

Fig. 2 is IP-MOEA algorithm flow chart

Fig. 3 is solar seawater desalination Interval Regression based on MatlabGUI and Optimization Platform;

Fig. 4 is the Interval Regression image display of GUI platform;

Fig. 5 is the interval multiple-objection optimization image display of GUI platform;

Fig. 6 is the Microsoft Excel of optimal solution set and desired value;

Fig. 7 be GUI platform seek ginseng display module;

Fig. 8 is that the Optimal Parameters of GUI platform arranges module;

Fig. 9 is the control knob module of GUI platform.

Concrete implementing measure

Below in conjunction with concrete accompanying drawing and example, the embodiment to institute of the present invention extracting method is described in detail.

1. set up the solar seawater desalination Interval Regression model based on LibSVM

(1) the choosing of training set and test set

In desalinization experiment, can experimental data be obtained, in order to training pattern in regression process, get the experimental data of 2/3rds as training set sample, using remaining data as test set pattern detection regression model.(2) interval of data is exported

In order to truly reflect the uncertainty in engineering reality, the present invention adopts interval expression to export, and this just needs to test the determination output interval obtained.Specifically, for a large amount of there are the data of identical input time, maximal value will be exported as the output interval upper bound, minimum value as output interval lower bound, for when not there are the data of identical input, original output is added the product of average area radius and Gauss disturbance as the output interval upper bound, original output deducts the product of average area radius and Gauss disturbance as interval lower bound.Concrete mode is such as formula shown in (3).

There are the data of identical input:

The upper bound=original output maximal value, (3)

After class=original output minimum value.

Not there are the data of identical input:

The upper bound=original output+| average area radius * Gauss disturbance |, (4)

Lower bound=original output-| average area radius * Gauss disturbance |.

(3) the choosing of support vector machine kind

LibSVM tool box provides five kinds of supporting vector machine models, is c-SVC, v-SVC, one-classSVM, ε-SVR, v-SVR respectively, is arranged by-s parameter.Wherein first three plants model for classification problem, and latter two model is used for regression problem.Present invention employs the foundation that latter two model and ε-SVR, v-SVR carry out the Interval Regression model of desalinization.

Wherein, the model of ε-SVR (ε-SupportVectorRegression) is:

$\underset{w,b,\mathrm{\ξ},{\mathrm{\ξ}}^{*}}{\min }\frac{1}{2}{w}^{T}w+C\underset{i=1}{\overset{l}{\Σ}}{\mathrm{\ξ}}_i+C\underset{i=1}{\overset{l}{\Σ}}{\mathrm{\ξ}}_i^{*}$

subjecttow ^Tφ(x _i)+b-z _i≤ε+ξ _i,(5)

$z_i-w^T\mathrm{\φ}\left(x_i\right)-b\≤\mathrm{\ϵ}+{\mathrm{\ξ}}_i^{*},$

${\mathrm{\ξ}}_i,{\mathrm{\ξ}}_i^{*}\≥o,i=1,2,\mathrm{...},l,$

Wherein, ξ _iwith be slack variable, C is penalty function coefficient, and w is Optimal Separating Hyperplane normal vector, and b is the constant term of Optimal Separating Hyperplane, φ (x _i) be x _ithrough transforming to the mapping of higher dimensional space, ε is insensitive function, z _iit is output variable.

The model of ν-SVR (ν-SupportVectorRegression) is

$\underset{w,b,\mathrm{\ξ},{\mathrm{\ξ}}^{*},\mathrm{\ϵ}}{\min }\frac{1}{2}{w}^{T}w+C(\mathrm{\υ}\mathrm{\ϵ}+\frac{1}{l}\underset{i=1}{\overset{l}{\Σ}}({\mathrm{\ξ}}_i+{\mathrm{\ξ}}_i^{*}\left)\right)$

subjecttow ^Tφ(x _i)+b-z _i≤ε+ξ _i,(6)

$z_i-w^T\mathrm{\φ}\left(x_i\right)-b\≤\mathrm{\ϵ}+{\mathrm{\ξ}}_i^{*},$

${\mathrm{\ξ}}_i,{\mathrm{\ξ}}_i^{*}\≥o,i=1,2,\mathrm{...},l,\mathrm{\ϵ}\≥0,$

Wherein, ξ _iwith be slack variable, C is penalty function coefficient, and w is Optimal Separating Hyperplane normal vector, and b is the constant term of Optimal Separating Hyperplane, φ (x _i) be x _ithrough transforming to the mapping of higher dimensional space, ε is insensitive function, z _iit is output variable.υ can control the number of support vector.

When selecting SVR type, the mode of optimizing is adopted to select.

(4) the choosing of support vector machine kernel function

Select different kernel functions, can generate different support vector machine, conventional kernel function has following four kinds:

A. linear kernel function K (x, y)=xy;

B. Polynomial kernel function K (x, y)=[(xy)+1] d;

C. radial basis function K (x, y)=exp (-| x-y|^2/d^2);

D. two layers of neural network kernel function K (x, y)=tanh (a (xy)+b).

Wherein, K is exactly kernel function, its role is to low dimension linearly inseparable sample sample can be divided to map to High-dimensional Linear.

Patent of the present invention is when arranging kernel function type, and same employing optimizing mode, finds optimum kernel function type.(5) structure of Interval Regression model

Build desalinization Interval Regression model and have following step:

Step1: tested by desalinization, obtains raw data, composing training collection data and test set data;

Step2: training set is exported data interval, obtains the interval upper bound and lower bound respectively;

Step3: training set data is normalized;

Step4: the circulation step-length that the penalty-c in LibSVM tool box, gamma function-g, kernel function type-t, n-fold cross-verification model-v and SVR types of models-s five are set;

Step5: the error calculating crosscheck by svmtrain, is stored in best_c, best_g, best_t, best_s respectively by-the c ,-g in least error situation ,-t ,-s parameter;

Step6: by best_c, best_g, best_t, best_s, has regional computer training set data and is set in svmtrain, and training obtains optimum upper bound model model_up, optimum lower bound model model_down;

Step7: by training set data respectively with optimum upper bound model model_up, optimum lower bound model model_down is set in svmpredict, the upper bound of respectively output regression prediction and lower bound;

Step8: in like manner, by test set data respectively with optimum upper bound model model_up, optimum lower bound model model_down is set in svmpredict, the upper bound of respectively output regression prediction and lower bound.

2. based on the desalinization Optimized model of IP-MOEA

(1) the interval Model for Multi-Objective Optimization of solar seawater desalination

For solar seawater desalination problem, at maintenance dry air flow, hot-air flow velocity, and when environment hot air temperature is constant, the fresh water productive rate of device depends primarily on soft air by the water capacity difference before and after condenser; Produce water cost to produce primarily of the power of condenser, therefore, adopt steam to be shown as this by the thermometer before and after condenser, more intuitively and be easy to measurement.In solar seawater desalination problem, due to the impact of the uncertain factors such as environment, make fresh water productive rate and produce water cost that there is certain uncertainty, thus, set up interval two objective optimization model of solar seawater desalination problem, and adopt the evolution optimization method IP-MOEA algorithm of the interval multi-objective optimization question of a kind of effective solution to solve.

Optimization problem is such as formula shown in (7):

minf(x)＝(1/f ₁(x),f ₂(x))，(7)

Wherein x=(x ₁, x ₂, x ₃, x ₄) ∈ X, x ₁for the water capacity difference before and after condenser, x ₂for the flow of dry air, x ₃for the flow velocity of hot-air, x ₄for the temperature of hot-air, f ₁x () is desalinization productive rate, f ₂x () is desalinization cost, shown in (8):

1/f ₁(x)＝[1/svmpredict(model ₁_up),1/svmpredict(model ₁_down)],(8)

f ₂(x)＝[svmpredict(model ₂_down),svmpredict(model ₂_up)],

Wherein, svmpredict () is Support vector regression function, model _i_ down and model _i_ up is respectively the lower bound and upper bound optimum regression model of training i-th target obtained.

(2) IP-MOEA algorithm

Consider following interval parameter multiple goal minimization problem:

minf(x,c)＝(f ₁(x,c),f ₂(x,c),…,f _m(x,c)),

$\begin{array}{cc}s.t.& x\∈S\⊆R^n\end{array},---\left(9\right)$

$c={(c_1,c_2,\mathrm{...},c_l)}^T,c_k=\[{\underset{\‾}{c}}_k,{\stackrel{\‾}{c}}_k\],k=1,2,\mathrm{...},l,$

Wherein, x is that n ties up decision variable, and S is the decision space of x; f _i(x, c), i=1,2 ..., m, be i-th objective function containing interval parameter, c is interval vector parameter, wherein, c _kfor a kth component of c, c _kwith be respectively c _klower limit and the upper limit.

In order to define the relation that is dominant based on interval, first Limbourg and Aponte define 2 interval order relations.

Consider a problem (9) 2 and separate x ₁and x ₂, i-th target function value of its correspondence is respectively f _i(x ₁, c) and f _i(x ₂, c), i=1,2 ..., m.The interval order relation provided is defined as follows: claim f _i(x ₁, c) under interval meaning, be less than f _i(x ₂, c), be designated as f _i(x ₁, c) < _iNf _i(x ₂, c), and if only if f _i(x ₁, lower limit c) and the upper limit are not more than f respectively _i(x ₂, lower limit c) and the upper limit, and these 2 intervals are unequal, that is:

$\begin{array}{c}{f}_i(x_1,c){<}_{IN}{f}_i(x_2,c)\&DoubleLeftRightArrow;\\ \underset{\&OverBar;}{f_i}(x_1,c)\&le;\underset{\&OverBar;}{f_i}(x_2,c)^{\stackrel{\&OverBar;}{f}}_i(x_1,c)\&le;{\stackrel{\&OverBar;}{f}}_i(x_2,c)^f_i(x_1,c)\&NotEqual;f_i(x_2,c),\end{array}---\left(10\right)$

Work as f _i(x ₁, c) < _iNf _i(x ₂, c) and f _i(x ₂, c) < _iNf _i(x ₁, when c) being all false, claim f _i(x ₁, c) under interval meaning with f _i(x ₂, c) not comparable, be designated as f _i(x ₁, c) || f _i(x ₂, c).

Then, based on above-mentioned interval order relation, the following interval that they define solution is further dominant relation: claim x ₁interval is dominant x ₂, be designated as x ₁> _iPx ₂, and if only if x ₁arbitrary objective function interval, all under interval meaning, be less than x ₂corresponding objective function is interval, or and x ₂corresponding objective function interval is not comparable, and at least there is x ₁an objective function interval, under interval meaning, be greater than x ₂corresponding objective function is interval, that is:

If x ₁neither interval is dominant x ₂, and x ₂also the not interval x that is dominant ₁, then x is claimed ₁and x ₂mutually not intervally to be dominant, to be designated as x ₁|| _iPx ₂.

The basic thought of IP-MOEA algorithm is: adopt NSGA-II normal form, realize the evolution of population, when the relation that is dominant of relatively Different Evolutionary individuality, utilize the above-mentioned relation that is dominant based on interval, replace traditional Pareto to be dominant relation, obtain that there is the non-by the individuality that is dominant of different sequence value.Concrete steps are as follows:

Step1: initialization scale is the population P (0) of N; Get evolutionary generation t=0;

Step2: be the genetic manipulation such as algorithm of tournament selection, crossover and mutation of 2 by scale, generate progeny population Q (t) of identical scale;

Step3: merge population P (t) and Q (t), and be denoted as R (t);

Step4: the relation that is dominant adopting the confidence level lower bound that to be dominant based on interval, asks for sequence value individual in R (t); Calculate the crowding with the individuality of identical sequence value; Choose top n advantage individual, form population P (t+1) of future generation;

Step5: whether decision algorithm end condition meets, if so, exports optimal solution set; Otherwise, make t=t+1, go to step 2.

The process flow diagram of IP-MOEA algorithm as shown in Figure 2.

3. build the desalinization Interval Regression based on MatlabGUI and optimal design platform

For the ease of parameters and display translation result, patent of the present invention realizes finally by visual desalinization Interval Regression and optimal design platform.GUI platform mainly comprises image output display module, LibSVM tool box parameter optimization display module, Optimal Parameters arrange module and control knob module, as shown in Figure 3.

The function of image output display module is output regression and the image after optimizing, and is divided into and returns image display and optimized image display module.Return image display and comprise productive rate change curve in training set, cost change curve in training set, productive rate change curve in test set, cost change curve in test set.As shown in Figure 4.Optimized image display module is the productive rate-cost leading surface figure of optimal solution set, as shown in Figure 5.Export the Excel form of optimal solution set and desired value, as shown in Figure 6.

The function of optimizing parameter display module is the optimizing parameter in display LibSVM tool box, is divided into the parameter display of productive rate and cost two aspects.Wherein, yield aspects comprises the display of the display of optimum bound punishment parameter-c parameter, the optimum display of bound gamma function-g parameter, the display of optimum bound kernel function-t parameter and optimum bound SVR type-s parameter.Equally, cost aspect comprises the display of the display of optimum bound punishment parameter-c parameter, the optimum display of bound gamma function-g parameter, the display of optimum bound kernel function-t parameter and optimum bound SVR type-s parameter.As shown in Figure 7.

The function that Optimal Parameters arranges module is the parameter arranging optimizing process necessity, comprises evolutionary generation, experiment number, population quantity, algorithm of tournament selection scale, crossover probability, mutation probability, and crossover and mutation profile exponent etc.As shown in Figure 8.

The function of control knob module triggers experiment process.Wherein, the function of the button that opens file opens experimental data, and load data; The function returning button triggers regression routine; The function optimizing button triggers regression routine, and prerequisite arranges in module at Optimal Parameters to arrange all parameters; The function of * button is for new procedures provides trigger button; The function emptying button resets all Optimal Parameters to arrange, reset optimizing parameter display; The function of exit button exits desalinization GUI platform.As shown in Figure 9.

Claims (4)

1. the interval supporting vector machine model of a solar seawater desalination problem, and to the data-optimized method for designing of input, it is characterized in that: based on LibSVM tool box, set up the Interval Regression model of desalinization and cost, and to productive rate and cost two target interval optimization problem, adopt the interval multi-target evolution optimization method of IP-MOEA to be optimized, find the optimal solution set of the input variables such as hot air temperature, finally set up visual desalinization recurrence, Optimization Platform, the method comprises:

(1) the solar seawater desalination Interval Regression model based on LibSVM tool box is set up

Solar seawater desalination problem relates generally to following parameter: hot air temperature, hot-air flow velocity, dry air flow, evaporator outlet temperature, evaporator outlet water capacity, condensator outlet temperature, and condensator outlet water capacity.Because internal relations is complicated, is not easy to utilize existing mathematical model to go to build its mapping relations, utilizes LibSVM tool box can set up the Interval Regression model of desalination rate and its dependent variable easily and efficiently.

(2) based on the desalinization optimal design of IP-MOEA

For solar seawater desalination problem, at maintenance dry air flow, hot-air flow velocity, and when environment hot air temperature is constant, the fresh water productive rate of device depends primarily on soft air by the water capacity difference before and after condenser.Produce water cost to produce primarily of the power of condenser, therefore, adopt steam to be shown as this by the thermometer before and after condenser, more intuitively and be easy to measurement.In solar seawater desalination problem, due to the impact of the uncertain factors such as environment, make fresh water productive rate and produce water cost that there is certain uncertainty, thus, set up interval two objective optimization model of solar seawater desalination problem, and adopt the evolution optimization method algorithm IP-MOEA of the interval multi-objective optimization question of a kind of effective solution to solve.

(3) the desalinization Interval Regression based on MatlabGUI and optimal design platform is built

In Interval Regression link, need the optimizing parameter showing LibSVM tool box, and the comparison diagram of output interval regression curve and raw data points; In the link of optimal design, need to be provided with and close the parameter such as evolutionary generation, and export the Microsoft Excel of the decision variable values such as the hot air temperature after optimizing and corresponding target function value.The GUI platform of final structure solar seawater desalination Interval Regression and optimal design.

2. the parameter that relates generally to of solar seawater desalination problem according to claim 1, set up the Interval Regression model based on LibSVM tool box, it is characterized in that, build the following step of model:

Step1: tested by desalinization, obtains raw data, composing training collection data and test set data;

Step2: training set is exported data interval, obtains the interval upper bound and lower bound respectively;

Step3: training set data is normalized;

Step4: the circulation step-length that the penalty-c in LibSVM tool box, gamma function-g, kernel function type-t, n-fold cross-verification model-v and SVR types of models-s five are set;

Step5: the error calculating crosscheck by svmtrain, is stored in best_c, best_g, best_t, best_s respectively by-the c ,-g in least error situation ,-t ,-s parameter;

Step6: by best_c, best_g, best_t, best_s, and there is regional computer training set data be set in svmtrain, training obtains optimum upper bound model model_up, optimum lower bound model model_down;

Step7: training set data is set in svmpredict with optimum upper bound model model_up, optimum lower bound model model_down respectively, the upper bound of output regression prediction respectively and lower bound.

Test set data are set in svmpredict with optimum upper bound model model_up, optimum lower bound model model_down by Step8: in like manner respectively, the upper bound of output regression prediction respectively and lower bound.

3. according to claim 1 in desalinization, fresh water productive rate and the decision factor producing water cost, set up and there are two regional computer objective optimization model, and utilize and solve based on the be dominant IP-MOEA algorithm of relation of interval, it is characterized in that: optimization problem is such as formula shown in (1):

minf(x)＝(1/f ₁(x),f ₂(x))(1)

Wherein x=(x ₁, x ₂, x ₃, x ₄) ∈ X, x ₁for the water capacity difference before and after condenser, x ₂for the flow of dry air, x ₃for the flow velocity of hot-air, x ₄for the temperature of hot-air, f ₁x () is desalinization productive rate, f ₂x () is desalinization cost, shown in (2):

1/f ₁(x)＝[1/svmpredict(model ₁_up),1/svmpredict(model ₁_down)],(2)

f ₂(x)＝[svmpredict(model ₂_down),svmpredict(model ₂_up)],

Wherein, svmpredict () is Support vector regression function, model _i_ down and model _i_ up is respectively the lower bound and upper bound optimum regression model of training i-th target obtained.

4., according to claim 1 about input parameter, the visual requirement of Output rusults, set up the desalinization Interval Regression based on MatlabGUI and Optimization Platform, it is characterized in that:

Based on the desalinization Interval Regression of gui interface and the major function of Optimization Platform be, realize man-machine interaction, can optimizing process parameter be set on platform, trigger experiment process, optimizing parameter in display Interval Regression, export training set, the fresh water productive rate of test set, the regression curve and the raw data comparison diagram that produce water cost, optimize after productive rate-cost forward position figure, the Microsoft Excel of output optimal solution set and its desired value.