CN106339573A - Artificial bee colony optimized rare-earth mine underground water total nitrogen concentration soft measurement method - Google Patents

Abstract

The invention discloses an artificial bee colony optimized rare-earth mine underground water total nitrogen concentration soft measurement method which adopts a support vector machine as a soft measurement model for total nitrogen concentration of underground water of a rare-earth mine, and a penalty factor C and a radial basis kernel parameter g of the support vector machine and a parameter epsilon are optimized and designed by using an adaptive artificial bee colony algorithm. In the adaptive artificial bee colony algorithm, search zoom factors are adaptively adjusted by using feedback information of an adaptive value, and information of an optimal individual of an adjacent domain and an optimal individual of a global domain is infused to a Gaussian variation strategy to generate new individuals adaptively. By adopting the method, the soft measurement precision of the total nitrogen concentration of the underground of the rare-earth mine can be improved.

Description

The rare-earth mining area underground water total nitrogen concentration flexible measurement method that artificial bee colony optimizes

Technical field

The present invention relates to underground water total nitrogen concentration fields of measurement, especially relate to the rare-earth mining area that a kind of artificial bee colony optimizes Underground water total nitrogen concentration flexible measurement method.

Background technology

The southern area of China is rich in ionic heavy rare earth resource, and heavy rare earth is indispensable during modern industry produces Raw material.At present, the exploitation of ion type rareearth resource commonly uses in_situ leaching technique.This technique is exactly in ion type rareearth Bore on mine and open some liquid injection holes, then by liquid injection hole, chemical solution is injected into inside mine.But it is dilute in ionic weight In the recovery process of soil resource, it is injected into these chemical solutions within mine and can gradually penetrate into underground, cause underground water Pollution.The pollution control of ion type rareearth mining area underground water is one requisite in ion type rareearth ore recovery process Business.

Total nitrogen concentration can reflect whether ion type rareearth mining area groundwater quality pollutes to a certain extent.In order to effectively Grasp the concentration of total nitrogen in the underground water of ion type rareearth mining area, one kind of the flexible measurement method based on intelligence computation is very attractive Means.Artificial bee colony algorithm is a kind of novelty intelligence computation method proposing in recent years, and it has been successfully applied to respectively Plant in engineering practice.But Traditional Man ant colony algorithm during the hard measurement optimizing rare-earth mining area underground water total nitrogen concentration often Exist and be easily trapped into local optimum, the not high shortcoming of hard measurement precision.

Content of the invention

The purpose of the present invention is to optimize rare-earth mining area underground water total nitrogen concentration hard measurement for Traditional Man ant colony algorithm When be easily trapped into local optimum, the not high shortcoming of hard measurement precision, the rare-earth mining area underground water that a kind of artificial bee colony optimizes is proposed Total nitrogen concentration flexible measurement method.The present invention can improve the precision of rare-earth mining area underground water total nitrogen concentration hard measurement.

Technical scheme: the rare-earth mining area underground water total nitrogen concentration flexible measurement method that a kind of artificial bee colony optimizes, Comprise the following steps:

Step 1, continuous zn days collection sampling of ground waters in the rare-earth mining area needing hard measurement, and to the underground water collecting Sample carries out detecting water quality index: nitrite nitrogen, nitrate nitrogen, total phosphorus, ammonia nitrogen, water temperature, ph value, dissolved oxygen, total nitrogen, will gather The groundwater quality achievement data arriving is as sample data set；Then to the rare-earth mining area groundwater quality index sample data collecting Collection is normalized, and the training dataset that front 70% is set to SVMs, 30% is set to test data afterwards Collection；

Step 2, user's initiation parameter, described initiation parameter includes hard measurement span number of days fd, Population Size Popsize, radius of neighbourhood nk, maximum does not improve number of times limit, maximum evaluation number of times max_fes；

Step 3, makes current evolution algebraically t=0, optimal design parameter number d=3 of SVMs, and makes and currently commenting Valency number of times fes=0；

Step 4, randomly generates population in search spaceWherein: individual subscript I=1,2 ..., popsize, andRepresent population p_tIn i-th individual, it randomly generates formula For:

$b_{i,j}^t={\mathrm{lb}}_j+rand(0,1)\×({\mathrm{ub}}_j-{\mathrm{lb}}_j)$

Wherein dimension subscript j=1,2,3；Illustrate 3 values needing optimal design parameter of SVMs, that is,It is the penalty factor c of SVMs,It is radial direction base nuclear parameter g of SVMs,Be SVMs not Parameter ε in sensitive loss function；Rand (0,1) is the function producing random real number between [0,1]；lb_jAnd ub_jIt is respectively and prop up Hold the lower bound of the search space needing optimal design parameter for j-th and the upper bound of vector machine；

Step 5, calculates population p_tIn each is individualAdaptive valueWherein individual subscript i=1,2 ..., Popsize, individualAdaptive valueComputational methods be: with individualityAs the training parameter of SVMs, so SVMs is concentrated from training data and is learnt afterwards, and the input variable of wherein SVMs is normalized one day dilute Native groundwater in mining area matter achievement data: nitrite nitrogen, nitrate nitrogen, total phosphorus, ammonia nitrogen, water temperature, ph value, dissolved oxygen, total nitrogen；? Hold the rare-earth mining area underground water total nitrogen concentration value after vector machine is output as normalized fd days；Calculate the supporting vector training Mean square error ne in test data set for the machine_i, then make individualAdaptive value

Step 6, makes population p_tIn each individuality do not improve number of timesWherein individual subscript i=1,2 ..., popsize；

Step 7, makes Evaluation: Current number of times fes=fes+popsize, and makes average factor mnu=0.5；

Step 8, preserves population p^tIn optimum individual best^t；

Step 9, employs honeybee execution adaptability search operation, specifically comprises the following steps that

Step 9.1, makes counter i=1, and makes zoom factor list scflist be sky；

Step 9.2, makes new individual

Step 9.3, with mnu as average, 0.1 for standard deviation produce a gaussian random real number grv, then order scaling because Sub- scf=grv × 2-1；

Step 9.4, randomly generates positive integer rd1 between [1, d]；

Step 9.5, randomly generates two unequal positive integers ri1 and ri2 between [1, popsize]；

Step 9.6, order

Step 9.7, calculates new individual u^tAdaptive value fit (u^t)；

Step 9.8, if new individual u^tThan individualityMore excellent, then grv is added in zoom factor list scflist；

Step 9.9, in individualityWith new individual u^tBetween execution selection operation more new individual b_i ^tDo not improve number of times

Step 9.10, makes counter i=i+1；

Step 9.11, if counter i is less than or equal to popsize, goes to step 9.2, otherwise goes to step 9.12；

Step 9.12, calculates the mean value meanscf of data in zoom factor list scflist；

Step 9.13, randomly generates a real number rw between [0.8,1.0]；

Step 9.14, makes mnu=rw × mnu+ (1-rw) × meanscf；

Step 9.15, goes to step 10；

Step 10, according to population p_tMiddle individual adaptive value calculates all individual select probability；

Step 11, observes honeybee according to population p_tIn each individual select probability select individual execution adaptability Gauss and become ETTHER-OR operation generate new individual, then execution selection operation and calculate individuality do not improve number of times, specifically comprise the following steps that

Step 11.1, makes counter i=1；

Step 11.2, according to population p_tIn each individual select probability individuality is gone out using roulette policy selectionAnd Make new individual

Step 11.3, makes neighborhood subscript rsi=(sei-nk+popsize) %popsize, wherein sei represent roulette plan Slightly select the subscript of individuality, % represents that complementation accords with；

Step 11.4, makes neighborhood optimum individualAnd make counter rt=1；

Step 11.5, makes neighborhood subscript rsi=(rsi+1) %popsize；

Step 11.6, if individualThan individual rsbest^tMore excellent, then makeOtherwise keep rsbest^tConstant；

Step 11.7, makes counter rt=rt+1；

Step 11.8, if rt is less than or equal to nk × 2, goes to step 11.5, otherwise goes to step 11.9；

Step 11.9, randomly generates positive integer rd2 between [1, d]；

Step 11.10, randomly generates positive integer rsn between [1, nk × 2], then makes random neighborhood subscript rni =(sei-nk+rsn+popsize) %popsize；

Step 11.11, makes average

Step 11.12, makes standard deviationWherein abs represents and takes absolute value Function；

Step 11.13, with grmean as average, grsd produces a gaussian random real number rval for standard deviation, if The value of rval is beyond [lb_rd2,ub_rd2] between scope, then gaussian random real number rval is regenerated using same method, Until the value of rval is without departing from [lb_rd2,ub_rd2] between scope；

Step 11.14, randomly generates positive integer ri3 between [1, popsize]；

Step 11.15, makes average

Step 11.16, makes standard deviationWherein abs represents the letter taking absolute value Number；

Step 11.17, with gbmean as average, gbsd produces a gaussian random real number bval for standard deviation, if The value of bval is beyond [lb_rd2,ub_rd2] between scope, then gaussian random real number bval is regenerated using same method, Until the value of bval is without departing from [lb_rd2,ub_rd2] between scope；

Step 11.18, randomly generates a real number rnw between [0,1]；

Step 11.19, order

Step 11.20, calculates new individual u^tAdaptive value fit (u^t), then in individualityWith new individual u^tBetween execute Selection operation, and calculate individualityDo not improve number of times

Step 11.21, makes counter i=i+1；

Step 11.22, if counter i is less than or equal to popsize, goes to step 11.2, otherwise goes to step 12；

Step 12, makes Evaluation: Current number of times fes=fes+popsize × 2；

Step 13, the search operation operator of execution search bee；

Step 14, preserves population p_tMiddle optimum individual best^t；

Step 15, makes current evolution algebraically t=t+1；

Step 16, repeat step 9 to step 15, until Evaluation: Current number of times fes reaches end after max_fes, will execute The optimum individual best obtaining in journey^tAs the training parameter of SVMs, then SVMs enters in training dataset Row training, by the rare-earth mining area underground water achievement data of normalized a day: nitrite nitrogen, nitrate nitrogen, total phosphorus, ammonia nitrogen, Water temperature, ph value, dissolved oxygen, total nitrogen are input to the SVMs training, and the output calculating SVMs gets final product hard measurement Go out total nitrogen concentration value after fd days for the rare-earth mining area underground water.

The present invention adopts SVMs as the soft-sensing model of rare-earth mining area underground water total nitrogen concentration, using adaptability Artificial bee colony algorithm carrys out ginseng in the penalty factor c of optimization design SVMs, radial direction base nuclear parameter g and insensitive loss function Number ε.In adaptability artificial bee colony algorithm, the feedback information using adaptive value adaptively adjusts search zoom factor, and Information fusion individual to neighborhood optimum individual and global optimum is adaptively produced new individual in Gaussian mutation strategy.This The bright hard measurement precision that can improve rare-earth mining area underground water total nitrogen concentration.

Brief description

Fig. 1 is the flow chart of the present invention.

Specific embodiment

Below by embodiment, and combine accompanying drawing, technical scheme is described in further detail.

Embodiment:

Step 1, determines and needs the Rare Earth Mine region of hard measurement, and in the region determining continuous zn=58 days locality under Water sample, and the sampling of ground water collecting is carried out detect water quality index: nitrite nitrogen, nitrate nitrogen, total phosphorus, ammonia nitrogen, water temperature, Ph value, dissolved oxygen, total nitrogen, using the groundwater quality collecting achievement data as sample data set；Then to the rare earth collecting Groundwater in mining area matter index sample data set is normalized, and the training data that front 70% is set to SVMs Collection, 30% is set to test data set afterwards；

Step 2, user's initiation parameter, described initiation parameter includes hard measurement span number of days fd=2, Population Size Popsize=50, radius of neighbourhood nk=5, maximum does not improve number of times limit=100, maximum evaluation number of times max_fes= 400000；

Step 3, makes current evolution algebraically t=0, optimal design parameter number d=3 of SVMs, and makes and currently commenting Valency number of times fes=0；

Step 4, randomly generates population in search spaceWherein: individual subscript I=1,2 ..., popsize, andRepresent population p_tIn i-th individual, it randomly generates formula For:

$b_{i,j}^t={\mathrm{lb}}_j+rand(0,1)\×({\mathrm{ub}}_j-{\mathrm{lb}}_j)$

Wherein dimension subscript j=1,2,3；Illustrate 3 values needing optimal design parameter of SVMs, that is,It is the penalty factor c of SVMs,It is radial direction base nuclear parameter g of SVMs,Be SVMs not Parameter ε in sensitive loss function；Rand (0,1) is the function producing random real number between [0,1]；Wherein lb=[0 0 0], Ub=[10,000 1 1], lb_jAnd ub_jIt is respectively needing for j-th under the search space of optimal design parameter of SVMs Boundary and the upper bound；

Step 5, calculates population p_tIn each is individualAdaptive valueWherein individual subscript i=1,2 ..., Popsize, individualAdaptive valueComputational methods be: with individualityAs the training parameter of SVMs, so SVMs is concentrated from training data and is learnt afterwards, and the input variable of wherein SVMs is normalized one day dilute Native groundwater in mining area matter achievement data: nitrite nitrogen, nitrate nitrogen, total phosphorus, ammonia nitrogen, water temperature, ph value, dissolved oxygen, total nitrogen；? Hold the rare-earth mining area underground water total nitrogen concentration value after vector machine is output as normalized 2 days；Calculate the supporting vector training Mean square error ne in test data set for the machine_i, then make individualAdaptive value

Step 6, makes population p_tIn each individuality do not improve number of timesWherein individual subscript i=1,2 ..., popsize；

Step 7, makes Evaluation: Current number of times fes=fes+popsize, and makes average factor mnu=0.5；

Step 8, preserves population p^tIn optimum individual best^t；

Step 9, employs honeybee execution adaptability search operation, specifically comprises the following steps that

Step 9.1, makes counter i=1, and makes zoom factor list scflist be sky；

Step 9.2, makes new individual

Step 9.3, with mnu as average, 0.1 for standard deviation produce a gaussian random real number grv, then order scaling because Sub- scf=grv × 2-1；

Step 9.4, randomly generates positive integer rd1 between [1, d]；

Step 9.5, randomly generates two unequal positive integers ri1 and ri2 between [1, popsize]；

Step 9.6, order

Step 9.7, calculates new individual u^tAdaptive value fit (u^t)；

Step 9.8, if new individual u^tThan individualityMore excellent, then grv is added in zoom factor list scflist；

Step 9.9, in individualityWith new individual u^tBetween execution selection operation more new individualDo not improve number of times

Step 9.10, makes counter i=i+1；

Step 9.11, if counter i is less than or equal to popsize, goes to step 9.2, otherwise goes to step 9.12；

Step 9.12, calculates the mean value meanscf of data in zoom factor list scflist；

Step 9.13, randomly generates a real number rw between [0.8,1.0]；

Step 9.14, makes mnu=rw × mnu+ (1-rw) × meanscf；

Step 9.15, goes to step 10；

Step 10, according to population p_tMiddle individual adaptive value calculates all individual select probability；

Step 11, observes honeybee according to population p_tIn each individual select probability select individual execution adaptability Gauss and become ETTHER-OR operation generate new individual, then execution selection operation and calculate individuality do not improve number of times, specifically comprise the following steps that

Step 11.1, makes counter i=1；

Step 11.2, according to population p_tIn each individual select probability individuality is gone out using roulette policy selectionAnd Make new individual

Step 11.3, makes neighborhood subscript rsi=(sei-nk+popsize) %popsize, wherein sei represent roulette plan Slightly select the subscript of individuality, % represents that complementation accords with；

Step 11.4, makes neighborhood optimum individualAnd make counter rt=1；

Step 11.5, makes neighborhood subscript rsi=(rsi+1) %popsize；

Step 11.6, if individualThan individual rsbest^tMore excellent, then makeOtherwise keep rsbest^tConstant；

Step 11.7, makes counter rt=rt+1；

Step 11.8, if rt is less than or equal to nk × 2, goes to step 11.5, otherwise goes to step 11.9；

Step 11.9, randomly generates positive integer rd2 between [1, d]；

Step 11.10, randomly generates positive integer rsn between [1, nk × 2], then makes random neighborhood subscript rni =(sei-nk+rsn+popsize) %popsize；

Step 11.11, makes average

Step 11.12, makes standard deviationWherein abs represents and takes absolute value Function；

Step 11.13, with grmean as average, grsd produces a gaussian random real number rval for standard deviation, if The value of rval is beyond [lb_rd2,ub_rd2] between scope, then gaussian random real number rval is regenerated using same method, Until the value of rval is without departing from [lb_rd2,ub_rd2] between scope；

Step 11.14, randomly generates positive integer ri3 between [1, popsize]；

Step 11.15, makes average

Step 11.16, makes standard deviationWherein abs represents the letter taking absolute value Number；

Step 11.17, with gbmean as average, gbsd produces a gaussian random real number bval for standard deviation, if The value of bval is beyond [lb_rd2,ub_rd2] between scope, then gaussian random real number bval is regenerated using same method, Until the value of bval is without departing from [lb_rd2,ub_rd2] between scope；

Step 11.18, randomly generates a real number rnw between [0,1]；

Step 11.19, order

Step 11.20, calculates new individual u^tAdaptive value fit (u^t), then in individualityWith new individual u^tBetween execute Selection operation, and calculate individualityDo not improve number of times

Step 11.21, makes counter i=i+1；

Step 11.22, if counter i is less than or equal to popsize, goes to step 11.2, otherwise goes to step 12；

Step 12, makes Evaluation: Current number of times fes=fes+popsize × 2；

Step 13, the search operation operator of execution search bee；

Step 14, preserves population p_tMiddle optimum individual best^t；

Step 15, makes current evolution algebraically t=t+1；

Step 16, repeat step 9 to step 15, until Evaluation: Current number of times fes reaches end after max_fes, will execute The optimum individual best obtaining in journey^tAs the training parameter of SVMs, then SVMs enters in training dataset Row training, by the rare-earth mining area underground water achievement data of normalized a day: nitrite nitrogen, nitrate nitrogen, total phosphorus, ammonia nitrogen, Water temperature, ph value, dissolved oxygen, total nitrogen are input to the SVMs training, and the output calculating SVMs gets final product hard measurement Go out rare-earth mining area underground water total nitrogen concentration value after 2 days.

Specific embodiment described herein is only explanation for example to present invention spirit.The affiliated technology of the present invention is led The technical staff in domain can be made various modifications or supplement or replaced using similar mode to described specific embodiment Generation, but the spirit without departing from the present invention or surmount scope defined in appended claims.

Claims (1)

1. the rare-earth mining area underground water total nitrogen concentration flexible measurement method that a kind of artificial bee colony optimizes is it is characterised in that include following Step:

Step 1, continuous zn days collection sampling of ground waters in the rare-earth mining area needing hard measurement, and the sampling of ground water collecting is entered Row detection water quality index: nitrite nitrogen, nitrate nitrogen, total phosphorus, ammonia nitrogen, water temperature, ph value, dissolved oxygen, total nitrogen, by collect Groundwater quality achievement data is as sample data set；Then the rare-earth mining area groundwater quality index sample data set collecting is entered Row normalized, and the training dataset that front 70% is set to SVMs, 30% are set to test data set afterwards；

Step 2, user's initiation parameter, described initiation parameter includes hard measurement span number of days fd, Population Size popsize, Radius of neighbourhood nk, maximum does not improve number of times limit, maximum evaluation number of times max_fes；

Step 3, makes current evolution algebraically t=0, optimal design parameter number d=3 of SVMs, and makes Evaluation: Current Number fes=0；

Step 4, randomly generates population in search spaceWherein: individual subscript i= 1,2 ..., popsize, andRepresent population p_tIn i-th individual, it randomly generates formula and is:

$b_{i,j}^t={\mathrm{lb}}_j+rand(0,1)\×({\mathrm{ub}}_j-{\mathrm{lb}}_j)$

Wherein dimension subscript j=1,2,3；Illustrate 3 values needing optimal design parameter of SVMs, that is,It is The penalty factor c of SVMs,It is radial direction base nuclear parameter g of SVMs,It is the insensitive of SVMs Parameter ε in loss function；Rand (0,1) is the function producing random real number between [0,1]；lb_jAnd ub_jBe respectively support to J-th of amount machine needs lower bound and the upper bound of the search space of optimal design parameter；

Step 5, calculates population p_tIn each is individualAdaptive valueWherein individual subscript i=1,2 ..., Popsize, individualAdaptive valueComputational methods be: with individualityAs the training parameter of SVMs, so SVMs is concentrated from training data and is learnt afterwards, and the input variable of wherein SVMs is normalized one day dilute Native groundwater in mining area matter achievement data: nitrite nitrogen, nitrate nitrogen, total phosphorus, ammonia nitrogen, water temperature, ph value, dissolved oxygen, total nitrogen；? Hold the rare-earth mining area underground water total nitrogen concentration value after vector machine is output as normalized fd days；Calculate the supporting vector training Mean square error ne in test data set for the machine_i, then make individualAdaptive value

Step 6, makes population p_tIn each individuality do not improve number of timesWherein individual subscript i=1,2 ..., popsize；

Step 7, makes Evaluation: Current number of times fes=fes+popsize, and makes average factor mnu=0.5；

Step 8, preserves population p^tIn optimum individual best^t；

Step 9, employs honeybee execution adaptability search operation, specifically comprises the following steps that

Step 9.1, makes counter i=1, and makes zoom factor list scflist be sky；

Step 9.2, makes new individual

Step 9.3, with mnu as average, the 0.1 gaussian random real number grv producing for standard deviation, then make zoom factor Scf=grv × 2-1；

Step 9.4, randomly generates positive integer rd1 between [1, d]；

Step 9.5, randomly generates two unequal positive integers ri1 and ri2 between [1, popsize]；

Step 9.6, order

Step 9.7, calculates new individual u^tAdaptive value fit (u^t)；

Step 9.8, if new individual u^tThan individualityMore excellent, then grv is added in zoom factor list scflist；

Step 9.9, in individualityWith new individual u^tBetween execution selection operation more new individualDo not improve number of times

Step 9.10, makes counter i=i+1；

Step 9.11, if counter i is less than or equal to popsize, goes to step 9.2, otherwise goes to step 9.12；

Step 9.12, calculates the mean value meanscf of data in zoom factor list scflist；

Step 9.13, randomly generates a real number rw between [0.8,1.0]；

Step 9.14, makes mnu=rw × mnu+ (1-rw) × meanscf；

Step 9.15, goes to step 10；

Step 10, according to population p_tMiddle individual adaptive value calculates all individual select probability；

Step 11, observes honeybee according to population p_tIn each individual select probability select individual execution adaptability Gaussian mutation behaviour Make generate new individual, then execution selection operation and calculate individuality do not improve number of times, specifically comprise the following steps that

Step 11.1, makes counter i=1；

Step 11.2, according to population p_tIn each individual select probability individuality is gone out using roulette policy selectionAnd make new Individual

Step 11.3, makes neighborhood subscript rsi=(sei-nk+popsize) %popsize, wherein sei represent that roulette strategy selects Select out the subscript of individuality, % represents that complementation accords with；

Step 11.4, makes neighborhood optimum individualAnd make counter rt=1；

Step 11.5, makes neighborhood subscript rsi=(rsi+1) %popsize；

Step 11.6, if individualThan individual rsbest^tMore excellent, then makeOtherwise keep rsbest^tNo Become；

Step 11.7, makes counter rt=rt+1；

Step 11.8, if rt is less than or equal to nk × 2, goes to step 11.5, otherwise goes to step 11.9；

Step 11.9, randomly generates positive integer rd2 between [1, d]；

Step 11.10, randomly generates positive integer rsn between [1, nk × 2], then makes random neighborhood subscript rni= (sei-nk+rsn+popsize) %popsize；

Step 11.11, makes average

Step 11.12, makes standard deviationWherein abs represents the function taking absolute value；

Step 11.13, with grmean as average, grsd produces a gaussian random real number rval for standard deviation, if rval Value is beyond [lb_rd2,ub_rd2] between scope, then gaussian random real number rval is regenerated using same method, until The value of rval is without departing from [lb_rd2,ub_rd2] between scope；

Step 11.14, randomly generates positive integer ri3 between [1, popsize]；

Step 11.15, makes average

Step 11.16, makes standard deviationWherein abs represents the function taking absolute value；

Step 11.17, with gbmean as average, gbsd produces a gaussian random real number bval for standard deviation, if bval Value is beyond [lb_rd2,ub_rd2] between scope, then gaussian random real number bval is regenerated using same method, until The value of bval is without departing from [lb_rd2,ub_rd2] between scope；

Step 11.18, randomly generates a real number rnw between [0,1]；

Step 11.19, order

Step 11.20, calculates new individual u^tAdaptive value fit (u^t), then in individualityWith new individual u^tBetween execute selection Operation, and calculate individualityDo not improve number of times

Step 11.21, makes counter i=i+1；

Step 11.22, if counter i is less than or equal to popsize, goes to step 11.2, otherwise goes to step 12；

Step 12, makes Evaluation: Current number of times fes=fes+popsize × 2；

Step 13, the search operation operator of execution search bee；

Step 14, preserves population p_tMiddle optimum individual best^t；

Step 15, makes current evolution algebraically t=t+1；

Step 16, repeat step 9 to step 15 is until Evaluation: Current number of times fes reaches_{max_fes}After terminate, will be in implementation procedure The optimum individual best arriving^tAs the training parameter of SVMs, then SVMs is trained in training dataset, By the rare-earth mining area underground water achievement data of normalized a day: nitrite nitrogen, nitrate nitrogen, total phosphorus, ammonia nitrogen, water temperature, ph Value, dissolved oxygen, total nitrogen are input to the SVMs training, and the output calculating SVMs can go out Rare Earth Mine by hard measurement Total nitrogen concentration value after fd days for area's underground water.