CN105718632A - Multi-target optimal management method for remediation of underground water with uncertainty - Google Patents

Abstract

The invention discloses a multi-target optimal management method for remediation of underground water with uncertainty. According to the method, target function random evaluation, a random Pareto domination concept and a random ecological niche fitness sharing method are imported into a population evolution operation of an EMOTS on the basis of a multi-target random tabu search algorithm PEMOTS of an elitism selection strategy. The PEMOTS inherits the global search advantage of the EMOTS, Latin Hypercube Sampling LHS is imported into an elitism selection strategy adopted by the EMOTS so as to generate neighborhood solutions, so that non-interior solutions obtained through the algorithm can be convergent to true solutions and can be uniformly distributed along a trade-off curve. The core difference between the PEMOTS and similar methods is adopting sequential Gaussian simulation SGSIM to reduce the uncertainty of water-bearing system parameters; and meanwhile, the random multi-target evolution operation is imported, the variability of the search of Pareto optimal solutions is reduced. The method disclosed in the invention is coupled with an underground water flow program MODFLOW and a solute transport program MT3DMS, so that relatively strong reliability and robustness are provided in the process of solving a multi-target management model for the pollution abatement of the underground water with uncertainty.

Description

A kind of uncertain underground water repairs multiple-objection optimization management method

Technical field

The present invention relates to a kind of uncertain underground water and repair multiple-objection optimization management method, belong to Hydrology and Water Resources field.

Background technology

More than 20 year of past, multi-objective Evolutionary Algorithm (Multi-objective evolutionary algorithm, MOEA) has become as The study hotspot in intelligence computation field, and be widely used in solving groundwater management model.Along with MOEA is at water The deep application of resources domain, model needs to consider the factors such as multivariate, Complex multi-target, and the uncertainty of aqueous system. In Groundwater optimal management model, uncertain systematic parameter can cause current and the random change of solute transfer state that model exports Change, and then cause the uncertain change of management objectives functional value.Definitiveness multi-objective Evolutionary Algorithm, as based on elite retention strategy The random tabu search algorithm of multiple target (Elitist Multi-Objective Tabu Search, EMOTS) be not suitable for solving not The Optimal management model determined.To this end, how to consider the uncertainty impact on Optimized model of analogue model, and how to seek Seeking the balance solution (Pareto optimal solution) met between uncertain target is the important content of subsoil water optimum management research field.

Conventional uncertain water resource optimization problem only considers the uncertainty of constraints or single target, uses averaging method, Evaluate individuality with the realization of one group of stochastic variable, calculate individual goal average value of a function, and carry out Evolution of Population operation, this Method considers the uncertainty of parameter, but relatively easy to object function Uncertainty Management, optimizes the reliability solved the strongest.

Summary of the invention

The technical problem to be solved is to provide a kind of uncertain underground water and repairs multiple-objection optimization management method, is one Plant the random tabu search algorithm of multiple target based on elite retention strategy (Probabilistic Elitist Multi-Objective Tabu Search, PEMOTS), the method has higher computational efficiency, ensure that simultaneously algorithm to search variability little, and surely Fixed reliable Pareto optimal solution.

The present invention solves above-mentioned technical problem by the following technical solutions:

The present invention provides a kind of uncertain underground water to repair multiple-objection optimization management method, specifically comprises the following steps that

Step 1, sets up analogue model, for portraying reparation place underground water head and solute concentration distribution over time and space；

Step 2, determines Optimal management model, sets up Optimized model；

Step 3, what employing order Gauss conditions simulation SGSIM generated infiltration coefficient realizes sample set, and for different conditions Count lnK field and analogue model are exported and carry out uncertainty analysis and risk assessment, be used for reducing model uncertainty and selection The lnK sample set of Optimized model management design；

Step 4, the PEMOTS Optimization Method multiobjective management problem balance of selection solves, concretely comprises the following steps:

(1) generation initial solution and corresponding neighborhood population:

First, random fashion is used to produce initial solution s₀, initialize elite table, candidate list and taboo list, with initial solution be Basic point, producing number based on Latin Hypercube Sampling LHS is N_TSNeighborhood population S^t _NS；

(2) object function haphazard evaluation:

Use noise genetic algorithm NGA, calculate population S one by one with the lnK sample set generated in step 3^t _NSIn each individuality Target function value, and add up expectation and the variance of the target function value of each individuality；According to target function value, use random Pareto controls sequence and random niche technique calculates individual Pareto ranking and the crowded angle value of individuality；Use ideal adaptation degree Functional value archival strategy records the solution searched first, and the solution of repeat search then directly invokes the solution information in functional value storehouse；

(3) random multi-target evolution:

1. the selection of subsolution is planted: compare neighborhood population S^t _NSElite table S with previous generation_E ^t-1Pareto controllability, by S^t _NSIn all Noninferior solution is considered as candidate seed disaggregation S₁；By S₁Add the candidate list S of previous generation_C ^t-1In, from S₁And S_C ^t-1Merging solution in select non- Inferior solution is as candidate seed disaggregation S in the present age₃；Last from S₃Two solutions of middle selection crowding distance maximum are as follow-on kind of subsolution S_S ^t+1。

The most more New Policy: by neighborhood population S in the present age^t _NSElite table S with previous generation_E ^t-1Merge, retain noninferior solution and form the present age Elite table S_E ^t, simultaneously by neighborhood population S in the present age^t _NSIn uncontrolled in the elite table S of previous generation_E ^t-1Solution, add previous generation candidate Table S_C ^t-1In, remove the controlled solution in candidate list and two kind subsolutions S selected the present age_S ^t+1, it is updated to candidate list S in the present age_C ^t；

(4) judge whether to meet stopping criterion

During optimization, if having reached maximum iteration time set in advance, or seed selection subsolution collection when searching certain generation For sky, candidate list is also empty simultaneously, then cannot find next kind subsolution, it is impossible to enter the search in next stage, stop also Output Pareto optimal solution set.

Step 5, exports optimum results, utilizes lnK sample set that Pareto optimal solution carries out Monte Carlo MC analysis, inspection The reliability of the Pareto optimal solution set of PEMOTS output.

As the further prioritization scheme of the present invention, in step 2, the object function of Optimized model is as follows:

$\begin{array}{cc}MIN& RC=\mathrm{\α}\underset{i=1}{\overset{N_w}{\mathrm{\Σ}}}\underset{i=1}{\overset{N_t}{\mathrm{\Σ}}}\end{array}\left|Q_{i,t}\right|\left(z_{gs}-h_{i,t}\right){\mathrm{\Δt}}_t+{\mathrm{\βN}}_S$

MAX MR (%)=(mass_end/mass₀)×100

Wherein, RC for minimizing treatment cost, Q_i,tIt is the flow that draws water of i-th mouthful of well t stress phase, △ t_tIt it is the t stress phase Duration, z_gsFor earth's surface elevation, h_i,tIt is the calculated water head of i-th mouthful of well t stress phase, z_gs-h_i,tIt is the pumping head of i-th mouthful of well, N_wAnd N_tIt is respectively preliminary election and administers well number and stress issue, N_SCounting for known conditions, α and β is cost coefficient and the acquisition of drawing water The cost coefficient of condition point K value；MR is for maximizing contaminant remaining percentage ratio, mass_endFor administering the week end of term total amount of pollutant, mass₀For the total amount of pollutant under original state.

As the further prioritization scheme of the present invention, in step 2, the constraints of Optimized model includes administering the constraint of well sum, water Head constraint, hydraulic gradient retrains, and pollutant levels retrain, water withdrawal traffic constraints and overall balance constraint.

As the further prioritization scheme of the present invention, step 5 utilize lnK sample set Pareto optimal solution is carried out Monte Carlo MC analyzes, and the reliability of the Pareto optimal solution set of inspection PEMOTS output, particularly as follows: utilize lnK sample set to step The Pareto optimal solution set obtained in 4 carries out Monte Carlo MC analysis, adds up the average of each solution, confidence level is 95% Uncertain interval upper boundary values and lower border value, the degree of closeness averagely solved with MC by comparing Pareto to solve, it is judged that PEMOTS is for solving the reliability of uncertain subsoil water pollution amelioration problem.

As the further prioritization scheme of the present invention, the Three dimensional finite difference that groundwater modeling is developed by US Geological Survey Program MODFLOW solves；Head that solute transport model solves based on MODFLOW and velocity field, utilize Zheng Chunmiao Modularity three-dimensional solute transfer program MT3DMS of research and development solves.

The present invention uses above technical scheme compared with prior art, has following technical effect that and optimizes at uncertain subsoil water In problem of management, computational efficiency and Pareto stability of solution determine the most important factor whether algorithm is suitable for, PEMOTS often Use individual archival strategy based on a small amount of random sample number, greatly reduce the amount of calculation of object function haphazard evaluation, simultaneously with Machine multi-target evolution ensures the Pareto optimal solution that to search variability little, highly reliable.To this end, PEMOTS is uncertain Underground water pollution is repaired in multiobjective management problem and is had broad application prospects.

Accompanying drawing explanation

Fig. 1 is PEMOTS flow chart.

Fig. 2 is Pareto control ordering chart under the conditions of non-determined.

Fig. 3 is that underground water pollution repairs place schematic diagram.

Fig. 4 is pollutant risk level and the variation tendency of the coefficient of variation in the case of different condition is counted.

Fig. 5 is preferable PAT system PEMOTS optimum results and Monte Carlo simulation analysis result.

Detailed description of the invention

Below in conjunction with the accompanying drawings technical scheme is described in further detail:

The present invention constructs the uncertain foggara reason model being made up of object function haphazard evaluation and random multi-target evolution and comments Valency method, i.e. based on elite retention strategy multiple target random tabu search algorithm (Probabilistic Elitist Multi-Objective Tabu Search, PEMOTS), and verified by infiltration coefficient uncertain two dimension subsoil water pollution amelioration application example. The method first application order Gauss conditions simulation (Sequential Gaussian Simulation, SGSIM) generates lnK and realizes sample Collection, and count lnK field and model are exported for different conditions and carry out uncertainty analysis and risk assessment, to reduce model Uncertain and selection is used for optimizing the lnK sample set of design.Then, introduce random Pareto and control sequence and random your pupil Border technology, selects the elite solution of population, intersects and the multi-target evolution operation such as variation, optimize and obtain Pareto optimal solution Collection.This invention is by PEMOTS and Simulation of Groundwater Flow program MODFLOW and solute transfer simulation program MT3DMS coupling Altogether, in order to design the multiple-objection optimization scheme of underground water pollution control under condition of uncertainty.

The program design of the random tabu search algorithm of multiple target based on elite retention strategy (PEMOTS) be object function with Machine is evaluated and random multi-target evolution two parts, as it is shown in figure 1, concrete solution procedure is as follows:

1. object function haphazard evaluation

Object function haphazard evaluation uses the thought of noise genetic algorithm (NGA) to process the uncertainty of model parameter.Due to There is uncertain factor in major part Actual Water Resource optimization problem, needs by one group of uncertainty realizing portraying mode input. With and need to solve two problems: use one group which type of realize portraying the uncertainty of numerical model？Realize from this group The great sample number of middle selection carrys out the object function that haphazard evaluation is individual？We be respectively adopted SGSIM DSMC and Individual archival strategy, solves two above problem.

1) DSMC of SGSIM: SGSIM is a kind of random mould order simulation thought and Gauss simulation combined Plan method.First all conditions data are carried out normal transformation, as priori conditions probability distribution, according to a stochastic simulation road Footpath, sequentially simulates each unconditional point value, is a little all modeled until needing to be estimated, and repeats above step simulation and generates appointment The realization of number constitutes stochastic variable and realizes sample set.The method can not only make the stochastic variable analogue value at condition point equal to real Measured value, it is also possible to maintain the spatial characteristics of variables model value to be distributed concordance with condition point measured value, be substantially reduced model defeated The uncertainty gone out.

2) individual archival strategy: in search procedure, for the individuality evaluated first, by this individuality (binary coding) and Object function statistical value (evaluating individual sample number, individual goal mean value functions and variance) leaves in individual archives, and Before calculating each individuality, it is first determined whether be stored in individual file store, if there is then increasing the sample number evaluating this individuality, And again add up target function value and update individual information in individual file store.The optimum individual occurred in some evolutionary process, Tournament selection is constantly retained when, it was demonstrated that its being dominant property of Pareto is strong, individual for this type of, by being continuously increased sample Number ensures stability of solution and reliability.Based on this, the sample set that will generate from SGSIM herein randomly selects a small amount of realization (5) evaluate and add up individual target function value.

2. random Multi-target evaluation

For uncertain multi-objective optimization question, the uncertainty of model parameter causes the uncertainty of object function.In target In function space, the solution that each decision vector is corresponding is not a fixing point, but a uncertain rectangular area；Non-it is subject to The Pareto forward position that control is deconstructed into is not a curve determined, but the range of indeterminacy being made up of Pareto up-and-down boundary, As shown in Figure 2.To this end, uncertain multi-target evolution is random Pareto to be controlled sequence be introduced into random niche technique In EMOTS, process the uncertainty of object function and seek Pareto optimal solution.

1) random Pareto controls sequence: for uncertain problem, it is judged that individual d₁Whether it is controlled by individual u1, needs individuality The difference of the average of target function value carries out parameter estimation and hypothesis testing.AssumeMeet t-to be distributed:

$T_i=\frac{\stackrel{\‾}{f_i}\left(d_1\right)-\stackrel{\‾}{f_i}\left(u_1\right)-(u_{d_1,i}-u_{u_1,i})}{\sqrt{\frac{S_{d_1,i}^2}{n}+\frac{S_{u_1,i}^2}{m}}}---\left(1\right)$

Wherein, T_iFor t-distribution function,WithIt is respectively individual d₁And u₁I-th object function Sample average and standard deviation,For individual d₁And u₁Expectation, n and m is for evaluating individual d₁And u₁Sample number, Namely the degree of freedom of this distribution is min (n-1, m-1).For given confidence level а, can table look-up and obtain corresponding quantile b (certainly It is n+m-2 by degree), obtainThe confidence interval that confidence level is а be

$\left(\stackrel{\‾}{f_i}\left(d_1\right)-\stackrel{\‾}{f_i}\left(u_1\right)-b\sqrt{\frac{S_{d_1,i}^2}{n}+\frac{S_{u_1,i}^2}{m}},\stackrel{\‾}{f_i}\left(d_1\right)-\stackrel{\‾}{f_i}\left(u_1\right)+b\sqrt{\frac{S_{d_1,i}^2}{n}+\frac{S_{u_1,i}^2}{m}}\}\right)---\left(2\right)$

Assume H_i:If the up-and-down boundary value of confidence interval be on the occasion of or negative value, it assumes that be false,OrIf the lower boundary of confidence interval is less than or equal to zero, coboundary is more than or equal to zero, at significance Assume under the conditions of horizontal а to set up.Each target is carried out parameter estimation and the significance test of object function average, if individual u₁Pareto is dominant d₁, thenAnd at least there is target j to makeBy the individuality two in population Two carry out random Pareto sequence, and individual Pareto ranking is equal to controlling its individual quantity in population, and Pareto is Excellent solution is not controlled by any individuality, and its ranking is 0.

2) random niche technique: niche technique is by judging between the Descartes's distance between individuality and microhabitat radius Relation, solves individual crowding, thus in the case of Pareto sequence is identical, selects the solution that crowding is little as advantage Body, to ensure the multiformity solved.With individual u₁And d₁As a example by, the first difference to the target function value of individual scaling transformationCarry out parameter estimation and hypothesis testing,For individual d₁And u₁Target function value. If assuming to set up, thenSame each object function average changed proportionally is carried out parameter estimation and Significance test, solves the Descartes's distance between individuality according to formula (3):

$E_{d_1{u}_1}=\sqrt{\underset{i=1}{\overset{k}{\Σ}}{\left(\stackrel{\‾}{O}\right(d_1)-\stackrel{\‾}{O}(u_1\left)\right)}^2}---\left(3\right)$

Wherein,For individual u₁And d₁Between Descartes's distance, k is the number of object function.

Individual crowding is solved by formula (4) and obtain, its be in this individuality and population other individuality the most spatially Microhabitat radius r_nicheInside solve out, r_nicheFor the meansigma methods of Descartes's distance between individuality two-by-two in population.

${\mathrm{nc}}_d=\left\{\begin{array}{cc}\underset{u=1}{\overset{N_P}{\Σ}}(1-\frac{E_{du}}{r_{niche}}),& ifd\≠uand{E}_{du}\≤r_{niche};\\ 0,& if{E}_{du}>r_{niche};\end{array}\right.---\left(4\right)$

Wherein, nc_dIt is the d individual individual crowding, N_PFor the individual sum in population, E_duBetween individual d and u Descartes's distance.

The present invention repairs multiobjective management problem for uncertain underground water, specifically comprises the following steps that

Step 1 sets up analogue model, for portray repair place underground water head and solute concentration over time and space point Cloth.Groundwater modeling can be solved by Three dimensional finite difference program MODFLOW that US Geological Survey develops, Head that solute transport model then solves based on MODFLOW and velocity field, the modularity three-dimensional utilizing Zheng Chun Seedling to research and develop is molten Matter migration program MT3DMS solves.Initial condition and boundary condition described in pattern of water flow and solute transport model are wanted Being consistent with actual place situation, the final analogue model set up is exactly pattern of water flow that is corrected and that check and solute transfer Model.

Step 2 determines Optimal management model, sets up Optimized model.Groundwater optimal management model is set up object function and setting Constraints is to set up the core of Optimal management model.This Study of The Underground water is repaired in multiobjective management problem, of paramount importance mesh Scalar functions is the minimum economic cost realizing recovery well optimization design on the basis of the various constraints of sewerage contentedly, in detail See that formula (5), recovery well optimization design include that well location puts the optimization with well yield.In addition residue dirt after underground water pollution is repaired The quality of dye thing has become the important indicator differentiating optimization design success or not, refers to formula (6) formula.

Object function one $\begin{array}{cc}MIN& RC=\mathrm{\α}\underset{i=1}{\overset{N_w}{\mathrm{\Σ}}}\underset{i=1}{\overset{N_t}{\mathrm{\Σ}}}\end{array}\left|Q_{i,t}\right|\left(z_{gs}-h_{i,t}\right){\mathrm{\Δt}}_t+{\mathrm{\βN}}_S---\left(5\right)$

Object function two MAX MR (%)=(mass_end/mass₀)×100 (6)

Wherein, RC for minimizing treatment cost, Q_i,tIt is the flow (m that draws water of i-th mouthful of well t stress phase³/d^-1), △ t_tIt is The duration (d) of t stress phase, z_gsFor earth's surface elevation (m), h_i,tIt is the calculated water head (m) of i-th mouthful of well t stress phase, z_gs-h_i,tIt is The pumping head (m) of i mouth well, N_wAnd N_tIt is respectively preliminary election and administers well number and stress issue, N_SCount for known conditions, α and β For cost coefficient and the cost coefficient of acquisition condition point K value of drawing water；MR is for maximizing contaminant remaining percentage ratio, mass_endFor Administer the week end of term total amount of pollutant (Kg), mass₀For the total amount of pollutant (Kg) under original state.

Constraints includes administering the constraint of well sum, and head retrains, and hydraulic gradient retrains, and pollutant levels retrain, water withdrawal Traffic constraints and overall balance constraint.The object function of change at random and constraints constitute uncertain groundwater remediation optimization The mathematical model of problem of management.

The uncertainty analysis of step 3 analog systems

This research use SGSIM condition simulation generate infiltration coefficient realize sample set.Literary composition utilize SGSIM method produce not LnK distribution in the case of counting with condition, produces respectively and realizes the sample set that number is 1000.Based on lnK sample set, utilize and cover The pollution concentration distribution after the match of the special different infiltration coefficient of Caro method simulation, uses the root-mean-square error of actual value and condition simulation value RMS carries out infiltration coefficient uncertainty analysis, and it is dirty to utilize pollution risk level and pollutant risk variation coefficient to carry out quantitative analysis Dye risk.In conjunction with the optimization treatment cost of pollution risk level and PAT system, select optimum condition to count, and used In the multi-objective optimization design of power of uncertain PAT system.

Step 4 is selected PEMOTS optimisation technique to solve the balance of multiobjective management problem and is solved.Optimization Solution technology is exactly in decision-making In the Feasible Solution Region of variable, by judging that target function value and constraints select uncontrolled decision variable to combine, as many mesh The compromise solution of mark optimization problem, the pros and cons that policymaker weighs between each target make management decision-making.

Step 5 exports optimum results, utilizes lnK sample set that PEMOTS optimizes the Pareto optimal solution obtained and carries out covering spy Caro is analyzed, add up the average of each solution, confidence level be 95% uncertain interval upper boundary values and lower border value, its In to obtain average solution by Monte Carlo simulation be the most reliable and the most stable.

Below by specific embodiment, technical scheme is further elaborated:

Below one uncertain groundwater remediation multiobjective management problem of design, utilizes PEMOTS optimisation technique to solve and meets management The balance solution of target and constraints.

2.1 PROBLEMSThe

One stochastic variable of this research design is the uncertain underground water repair system of infiltration coefficient (K).Hydrology ground, study area The pollution situation of matter condition and nitrate (in terms of nitrogen) is as shown in Figure 3.In figure, "●" is 4 mouthfuls of pretreatment wells, the rectangle of Lycoperdon polymorphum Vitt Region is concentration restriction range, i.e. optimizes design by PAT and carrys out the concentration higher limit less than regulation of control area.Water-bearing layer Infiltration coefficient be 2.5m/d for meeting average, variance is the lnK field of the random distribution of 0.2, infiltration coefficient uncertain, leads Cause the change at random of model delivery head and pollutant levels.

The uncertainty analysis of 2.2 analog systemss

This research use SGSIM condition simulation generate infiltration coefficient realize sample set.Literary composition is provided with 6 kinds of situations, condition point Number is respectively N_S=10,20,30,40,50,60, utilize SGSIM method to produce the lnK distribution in the case of different condition is counted, Produce respectively and realize the sample set that number is 1000.Based on lnK sample set, utilize Monte-carlo Simulation difference infiltration coefficient field Under pollution concentration distribution, use the root-mean-square error (RMS) of actual value and condition simulation value to carry out infiltration coefficient uncertainty and divide Analysis, and utilize pollution risk level and pollutant risk variation coefficient to carry out quantitative analysis pollution risk, as shown in Figure 4.From Fig. 4 Can be seen that the increase counted along with condition, pollution risk reduces and tends towards stability value, it is considered to the optimization of PAT system is administered into This, select 40 condition points to generate the sample set that 1000 lnK realize herein, and use it for uncertain PAT system In multi-objective optimization design of power, seek random Pareto optimal solution.

2.3 application PEMOTS optimisation techniques

PEMOTS relevant parameter of the present invention be provided that computer algebra is 100；Population Size is 100；Pareto disaggregation mistake Filter size is 100；The discretization interval number of each parameter (pump-out) is 32；Uniform crossover probability is 0.95；Single-point becomes Different probability is 0.05；M ü hlenbein mutation probability is 0.25；The subsample number of evaluation objective functional value is 5, for having searched The individuality that rope is crossed, often re-searches for once, and the subsample number evaluating this individuality increases by 5 (maximum subsample number is 30).Using Penalty function is often used the administrative model of constraint to be converted to unconfinement administrative model, i.e. to not when intelligent algorithm is optimized design Meet the solution of constraints, use the form of penalty function to be added in the object function of correspondence.Use PEMOTS optimum results such as figure Shown in 5.

The relative analysis of 2.4 optimum results

Herein based on 40 equally distributed condition points, SGSIM condition simulation is utilized to generate the sample set that 1000 lnK realize, And from this sample set, randomly select 5 target function values realizing originally evaluating individuality as increment, by the many mesh of PEMOTS Mark is evolved and is found uncertain Pareto optimal solution, as shown in Figure 5 ("+" solve).For the reliability of test result, use 1000 The sample set of individual realization carries out Monte Carlo simulation analysis to each Pareto solution, adds up the average ("+" solve) of each solution, and puts Menstruation equals upper boundary values ("○" solution) and the lower border value (" △ " solves) in the uncertain interval being 95%, wherein passes through Meng Teka It is the most reliable and the most stable that sieve simulation obtains average solution, and PEMOTS optimizes the Pareto optimal solution obtained and is distributed in the range of indeterminacy In, can judge in terms of following two for uncertain Pareto optimal solution superiority-inferiority, (1) single Pareto solves not Definitiveness interval width is the least, shows that the variability of this solution is little, and in practical problem, the variability of solution is little, then correspondence is reliable Property is strong；(2) Pareto angle distribution solves closer to average, and the reliability of solution is the strongest.The Pareto optimal solution that PEMOTS tries to achieve is close Solve in average, show that random Pareto controls sequence and random niche technique and processing the effectiveness of uncertain multiobjective selection, The Pareto optimal solution variability obtained is little, highly reliable.

The above, the only detailed description of the invention in the present invention, but protection scope of the present invention is not limited thereto, and any ripe Know the people of this technology in the technical scope that disclosed herein, it will be appreciated that the conversion expected or replacement, all should contain in the present invention Comprise within the scope of, therefore, protection scope of the present invention should be as the criterion with the protection domain of claims.

Claims (5)

1. a uncertain underground water repairs multiple-objection optimization management method, it is characterised in that specifically comprise the following steps that

Step 1, sets up analogue model, for portraying reparation place underground water head and solute concentration distribution over time and space；

Step 2, determines optimum management target, sets up Optimized model；

Step 3, what employing order Gauss conditions simulation SGSIM generated infiltration coefficient realizes sample set, and for different conditions Count lnK field and analogue model are exported and carry out uncertainty analysis and risk assessment, be used for reducing model uncertainty and selection The lnK sample set of Optimized model management design；

Step 4, the PEMOTS Optimization Method multiobjective management problem balance of selection solves, concretely comprises the following steps:

(1) generation initial solution and corresponding neighborhood population:

First, random fashion is used to produce initial solution s₀, initialize elite table, candidate list and taboo list, with initial solution be Basic point, producing number based on Latin Hypercube Sampling LHS is N_TSNeighborhood population S^t _NS；

(2) object function haphazard evaluation:

Use noise genetic algorithm NGA, calculate population S one by one with the lnK sample set generated in step 3^t _NSIn each individuality Target function value, and add up expectation and the variance of the target function value of each individuality；According to target function value, use random Pareto controls sequence and random niche technique calculates individual Pareto ranking and the crowded angle value of individuality；Use ideal adaptation degree Functional value archival strategy records the solution searched first, and the solution of repeat search then directly invokes the solution information in functional value storehouse；

(3) random multi-target evolution:

1. the selection of subsolution is planted: compare neighborhood population S^t _NSElite table S with previous generation_E ^t-1Pareto controllability, by S^t _NSIn all Noninferior solution is considered as candidate seed disaggregation S₁；By S₁Add the candidate list S of previous generation_C ^t-1In, from S₁And S_C ^t-1Merging solution in select non- Inferior solution is as candidate seed disaggregation S in the present age₃；Last from S₃Two solutions of middle selection crowding distance maximum are as follow-on kind of subsolution S_S ^t+1。

The most more New Policy: by neighborhood population S in the present age^t _NSElite table S with previous generation_E ^t-1Merge, retain noninferior solution and form the present age Elite table S_E ^t, simultaneously by neighborhood population S in the present age^t _NSIn uncontrolled in the elite table S of previous generation_E ^t-1Solution, add previous generation candidate Table S_C ^t-1In, remove the controlled solution in candidate list and two kind subsolutions S selected the present age_S ^t+1, it is updated to candidate list S in the present age_C ^t；

(4) judge whether to meet stopping criterion

During optimization, if having reached maximum iteration time set in advance, or seed selection subsolution collection when searching certain generation For sky, candidate list is also empty simultaneously, then cannot find next kind subsolution, it is impossible to enter the search in next stage, stop also Output Pareto optimal solution set；

Step 5, exports optimum results, utilizes lnK sample set that Pareto optimal solution carries out Monte Carlo MC analysis, inspection The reliability of the Pareto optimal solution set of PEMOTS output.

A kind of uncertain underground the most according to claim 1 water repairs multiple-objection optimization management method, it is characterised in that In step 2, the object function of Optimized model is as follows:

$\begin{array}{cc}MIN& RC=\mathrm{\α}\underset{i=1}{\overset{N_w}{\Σ}}\underset{t=1}{\overset{N_t}{\Σ}}\left|Q_{i,t}\right|(z_{gs}-h_{i,t}){\mathrm{\Δt}}_t+{\mathrm{\βN}}_S\end{array}$

MAX MR (%)=(mass_end/mass₀)×100

Wherein, RC for minimizing treatment cost, Q_i,tIt is the flow that draws water of i-th mouthful of well t stress phase, △ t_tIt it is the t stress phase Duration, z_gsFor earth's surface elevation, h_i,tIt is the calculated water head of i-th mouthful of well t stress phase, z_gs-h_i,tIt is the pumping head of i-th mouthful of well, N_wAnd N_tIt is respectively preliminary election and administers well number and stress issue, N_SCounting for known conditions, α and β is cost coefficient and the acquisition of drawing water The cost coefficient of condition point K value；MR is for maximizing contaminant remaining percentage ratio, mass_endFor administering the week end of term total amount of pollutant, mass₀For the total amount of pollutant under original state.

A kind of uncertain underground the most according to claim 1 water repairs multiple-objection optimization management method, it is characterised in that In step 2, the constraints of Optimized model includes administering the constraint of well sum, and head retrains, and hydraulic gradient retrains, pollutant levels Constraint, water withdrawal traffic constraints and overall balance constraint.

A kind of uncertain underground the most according to claim 1 water repairs multiple-objection optimization management method, it is characterised in that Step 5 utilize lnK sample set Pareto optimal solution carries out Monte Carlo MC analysis, the Pareto of inspection PEMOTS output The reliability of optimal solution set, particularly as follows: utilize lnK sample set that the Pareto optimal solution set obtained in step 4 is carried out Meng Teka Sieve MC analyzes, add up the average of each solution, confidence level be 95% uncertain interval upper boundary values and lower border value, The degree of closeness averagely solved with MC by comparing Pareto to solve, it is judged that PEMOTS is used for solving uncertain subsoil water pollution amelioration The reliability of problem.

A kind of uncertain underground the most according to claim 1 water repairs multiple-objection optimization management method, it is characterised in that Three dimensional finite difference program MODFLOW that groundwater modeling is developed by US Geological Survey solves；Solute transfer Head that model solves based on MODFLOW and velocity field, utilize the modularity three-dimensional solute transfer program that Zheng Chun Seedling is researched and developed MT3DMS solves.