CN103473465A - Method for optimizing spatial layout of land resources based on multi-target artificial immunization system - Google Patents

Abstract

The invention provides a method for optimizing the spatial layout of land resources based on a multi-target artificial immunization system. The method comprises the following steps: storing the evaluation result of the land suitability of a planned region in a base year by a grid file; encoding namely mapping the spatial layout scheme of the land resources into the antibodies of the artificial immunization system through integer coding, and representing one actual plot of land by one grid unit, which corresponds to one gene bit in each artificial antibody, in the planned region; setting a target function, setting constraint conditions, initiating an antibody population, calculating the target value vector quantities of the antibodies, then iterating, carrying out cloning, mutation, calculation on the target value vector quantities of the antibodies, population updating and judgment on terminal conditions until iteration is over, and decoding to obtain a Pareto optimizing scheme.

Description

Land resource space layout optimization method based on the multiple goal artificial immune system

Technical field

The invention belongs to the land resource space layout and distribute technical field rationally, particularly relate to a kind of land resource space layout optimization method based on the multiple goal artificial immune system.

Background technology

The space layout optimization of land resource is the important component part of land use planning work; its essence is under the constraint of regional planning phase land resource quantitative structure; nature (pH value, soil thickness etc.), society's (ownership, position condition), economic (price, economic benefit etc.) attributive character according to soil self; determine the purposes in each piece soil; the suitability utilized to improve Land in Regional Land, make the space layout of Area Land Resources more be conducive to Production&Operations Management and ecological environmental protection.Therefore, the space layout optimization problem of land resource is the multiple goal combinatorial optimization problem of a class Problem with Some Constrained Conditions from seeing in essence.

With respect to traditional multiple goal combinatorial optimization problem, the space layout optimization allocation of land resource is more complicated, solve difficulty larger: not only need to process traditional mathematics restriction relation, also must process complicated space constraint relation, and decision variable numerous (ten hundreds of).Thereby, due to the restriction of technical conditions, in land use planning in early days, the space layout of land resource mainly relies on all kinds of Land Use Divisions of delimitation on the artificial subjective space of planning expert to be achieved.The subject matter that relies on the expert to carry out subregion to the soil utilization is that the subregion result is too subjective, the partition scheme out of true, makes traditional land use planning scheme finally can't implement in actual soil utilization.On the other hand, along with the growth of China's population, the quick propelling of urbanization process, the irrational utilization of land resource is day by day obvious, it is too fast that emphasis shows as the expansion of city-building land scale, farming land (especially ploughing) runs off serious, people ground contradiction is sharp-pointed, and has caused serious ecosystem environment disaster.The action by government that traditional land use planning is considered to be is a kind of " strong politics, weak technology ", can not meet the primary demand of new period soil sustainable use and ecological environmental protection.Therefore, China's new round overall plan for land use work carry out with implementation process in, for the land use planning of villages and small towns level, concrete purposes that must clear and definite each piece soil.As can be seen here, the method for the subjective definite area land use of traditional dependence expert can not meet the actual demand of new period land use planning work.

In recent years, along with the fast development of Intelligent Optimization Technique and Spatial Information Technology, and the tight demand of China's land use planning work, some new theories and new method start to be widely used in land use planning work.Distribute field rationally in the land resource space layout, the intelligent optimization algorithm that relevant researcher starts based on different is both at home and abroad explored the space layout optimisation technique of land resource, and obtained effect preferably, the Optimized model related documents based on genetic algorithm for example: [1] Xi Yifan, Yang Maosheng, Shang Yaohua. the application [J] of genetic algorithm in the urban land function configuring. Northwest Building Engineering College's journal (natural science edition), 2001, 18 (04): the 190-194. related documents: [2] Zhang Honghui, Zeng Yongnian, Liu Huimin. the multiple goal soil utilizes Spatial optimum allocation model and application [J] thereof. Central South University's journal (natural science edition), 2011, 42 (04): the 1056-1065. related documents: [3] CAO K, YE X.Coarse-grained parallel genetic algorithm applied to a vector based land use allocation optimization problem:the case study of Tongzhou Newtown, Beijing, China[J] .StochEnv Res Risk A, 2012, 1-10., Optimized model related documents based on simulated annealing: [4] SANTE-RIVEIRA I, BOULLON-MAGAN M, CRECENTE-MASEDA R, et al.Algorithm based on simulated annealing for land-use allocation[J] .ComputGeosci, 2008, 34 (3): the 259-268. related documents: [5] AERTS J, VAN HERWIJNEN M, STEWART T J.Using simulated annealing and spatial goal programming for solving a multi site land use allocation problem[M] .Berlin, Springer-Verlag Berlin.2003:448-463., the Optimized model related documents based on particle swarm optimization: [6] LIU Y L, LIU D F, LIU Y F, et al.Rural land use spatial allocation in the semiarid loess hilly area in China:Using a Particle Swarm Optimization model equipped with multi-objective optimization techniques[J] .Sci China-Earth Sci, 2012,55 (7): 1166-1177. etc.

The proposition of above intelligent optimization model, advanced the land resource space layout to distribute relevant work and research rationally technically greatly, and the land resource space layout that is is towards robotization, intelligent direction development.Yet, in existing optimisation technique, also there is certain deficiency, main manifestations is:

(1) the space search ability of model a little less than

Intelligent optimization algorithm obtains optimum solution by random search, the mode that iterates mostly, and its counting yield sharply descends along with increasing of decision variable, and this phenomenon is also referred to as " dimension disaster ".Along with the increase of ground number of blocks, existing soil utilizes the problem of the space search scarce capacity of space layout Optimized model also to highlight thereupon, the situation that algorithm is not restrained even occurs.The problems referred to above that intelligent algorithm exists show, with the way of search of randomly changing ground block type, can not meet the demand that extensive soil utilizes the space layout Optimizing Search.Therefore, must optimize from the land resource space layout fundamental characteristics of this space optimization problem, the space search strategy of improved model, the search efficiency of enhancing algorithm, improve convergence of algorithm speed and counting yield.

(2) the multi-objective method of model can not meet the needs that soil utilizes multiobjectives decision to support

The Optimum allocation of land use problem is the multi-objective optimization question of a class Problem with Some Constrained Conditions.In land resource Spatial optimum allocation model investigation field, most research has all adopted multiple goal is processed based on linear weight sum method.The advantage of the method is to calculate simple, yet defect is also apparent, as target weight be difficult to determine, very responsive to the shape of Pareto front end, can not process the recess of its front end etc.Therefore, needing to introduce more advanced multi-objective method is improved land resource space layout Optimized model.

Artificial immune system (Artificial Immune System, AIS) is the intelligent optimization algorithm that the various principles of an analoglike natural immune system are developed with mechanism.With respect to genetic algorithm, the intelligent optimization algorithms such as simulated annealing, particle swarm optimization and ant group algorithm, artificial immune system has its unique advantage at aspects such as optimizing ability, speed of convergence and maintenance diversity in the mid-autumn, thereby is widely used in a plurality of fields such as information security, machine learning, data mining and intelligent optimization.Yet artificial immune system does not also obtain enough attention in the land resource space layout field of distributing rationally at present, yet there are no relevant report both at home and abroad.Therefore, above-mentioned deficiency for existing Optimized model, the NICA algorithm that the present invention will be take in the multiple goal artificial immune system models is basis, for the basic characteristics of land resource space layout problem with solve demand, propose that a kind of space search ability is strong, novel towards land resource space layout optimisation technique based on the Pareto dominant strategy, realize robotization and the intelligent allocation of land resource space layout.

Summary of the invention

For the limitation of existing land resource space layout optimisation technique, invent the land resource space layout optimisation technique based on the multiple goal artificial immune system of a kind of intellectuality, robotization.For land use planning provides science, reasonable, accurate, practicable land resource space layout prioritization scheme, so for realize land resource rationally, efficient, sustainable use provides important leverage.

Technical scheme provided by the invention is a kind of land resource space layout optimization method based on the multiple goal artificial immune system, comprises the following steps:

Step 1, by the planning region appraisal of land suitability as a result figure adopt raster file to be stored;

Step 2, coding, encode a kind of land resource space layout scheme by two-dimensional integer, be mapped as the antibody of artificial immune system; Comprise with a grid cell in planning region and represent an actual plot, a gene position in corresponding artificial antibody, the characteristic information of locus ranks that each gene position has comprised corresponding plot number and ground class, described characteristic information comprises the land used type information to the configuration of the suitability score value of all ground class and current plot, the suitability score value according to appraisal of land suitability as a result the raster file of figure obtain;

Step 3, objective function setting, comprise two optimization aim, is respectively maximum suitability degree S and maximum compactness Comp,

$S=\mathrm{Max}\left({\mathrm{\Σ}}_{i=1}^N{\mathrm{Suit}}_i\right)---\left(1\right)$

In formula, N means the sum of grid cell in planning region, Suit _iit is the suitability score value of the ground class that i grid cell is current configured to it;

$\mathrm{Comp}=\mathrm{Min}\left({\mathrm{\Σ}}_{i=1}^n{\mathrm{LSI}}_j\right)---\left(2\right)$

In formula, the value that Comp is maximum compactness target, the number that n is patch in planning region, on space, grid cell adjacent and that the land used type is identical forms patch, LSI _jit is the shape index of j patch;

Step 4, constraint condition setting, comprise that setting land resource quantitative structure corresponding to land resource space layout scheme must equal the land resource quantitative structure that regional planning is set;

Step 5, antibody population initialization, establishing the chromosome population scale is AN, in conjunction with land resource quantitative structure project period, takes random mode to produce AN initial antibodies;

Step 6, antibody desired value vector calculation, comprise that employing formula (1) and formula (2) calculate the desired value vector of each antibody, and carry out the Pareto sequence, obtains Pareto and optimize disaggregation, as initial memory antibody population;

Step 7, clone, comprise current memory antibody copied to C part according to default clone's coefficient C, forms new antibody population;

Step 8, variation, comprise the antibody population that traversal is new, and each antibody is implemented to mutation operation;

Step 9, antibody desired value vector calculation, employing formula (1) and formula (2) are calculated the desired value vector of each antibody.

Step 10, population are upgraded, and antibody new population and original memory antibody population are merged, and the population after being combined re-start the Pareto sequence, obtain new Pareto disaggregation, obtain new memory antibody population;

The judgement of step 11, end condition, reach default maximum iteration time G if calculate the current iteration number of times, termination of iterations, and decoding obtains the Pareto prioritization scheme.

And the shape index LSI of patch obtains as follows,

$\mathrm{LSI}=\frac{P}{4\sqrt{A}}---\left(3\right)$

In formula, P means the girth of patch, and A means the area of patch.

And the performing step of step 5 is as follows,

Step 5.1, build list Y, the length of Y be the number of class, each element representation needed grid cell number of class accordingly in Y;

Step 5.2, produce empty antibody, and the value of each gene position of antibody is empty;

Step 5.3, travel through each gene position, and for current gene position, land suitability score value based on plot builds wheel disc, adopts the roulette strategy to determine at random its ground class, the principle that wheel disc builds as shown in the formula,

$p_l=\frac{f_l}{\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{f}_l}---\left(5\right)$

In formula, p _lfor the selected probability of ground class l, f _lfor the current plot suitability score value of class l over the ground, L is the sum of class accordingly in list Y; A random random number rd that value is 0-1, the judgement position of rd in wheel disc, and then the ground class L in definite current plot of producing _i, and by ground class L _ivalue corresponding in list Y deducts 1, if the element value subtracted after 1 is 0, this element is deleted from list Y;

Step 5.4, repeating step 5.3, until list Y is empty, complete the generation of initial antibodies.

And the gene position i of traversal antibody, take grid cell as picture dot, establishes gene position and add up to N, the initial value of i is 0; The performing step of step 8 is as follows,

Step 8.1, i=i+1, the ground class of current gene position is L _i, generate the random number rnd of value between 0-1, if rnd is less than default antibody variation rate P _m, enter next step; Otherwise continue to repeat this step and jump to next gene position;

Step 8.2, obtain the neighborhood picture dot collection P of current gene position, and add up ground class and the corresponding picture dot number comprised in P;

Step 8.3, if only comprise a kind of ground class in neighborhood P, return to step 8.1, otherwise enter next step;

Step 8.4, various places class L in statistics P _jprobability of occurrence, be designated as probability P _j;

Step 8.5, obtain the suitability score value S of the corresponding plot of current gene position to all types land used _j;

Step 8.6, the comprehensive selection probability C of calculating various places class _j=P _j* S _j;

Step 8.7, according to the comprehensive selection probability of various places class, adopt ground class L of the random selection of roulette strategy _cground class after implementing to make a variation as current gene position;

Step 8.8 searches out a ground class and is encoded to L in antibody _cgene position C, with current gene position exchange ground class coding;

Step 8.9, if i<N returns to step 8.1 and continues to carry out, otherwise exits circulation, antibody variation has operated.

And step 8.8 comprises the following steps,

Step 8.1.1, a position y of the random generation in the zone in antibody except current gene position;

Step 8.1.2, if the ground class L of position y _yfor L _c, enter next step, otherwise return to step 8.1.1;

Step 8.1.3, obtain the neighborhood picture dot collection P of position y, and add up the quantity of ground class in P, if only have a kind of ground class in P, returns to step 8.1.1, otherwise enter next step;

Step 8.1.4, if comprise L in the neighborhood picture dot of y _iloop termination, using position y as antibody gene exchange position, otherwise return to step 8.1.1, continues to search the exchange position.

And, in step 5, for new Pareto disaggregation, the quantity TP that statistics Pareto separates, if TP is greater than default maximum Pareto and separates quantity MP,, according to the crowding distance size of each solution, delete TP-MP the Pareto that crowding distance is less and separate, remaining Pareto is deconstructed into new memory antibody population.

And the crowding distance computing method that each Pareto separates corresponding antibodies are as follows:

1) according to the desired value antagonist population of k optimization aim, sorted;

2) sort the most front and antibody most end is set to a positive infinity value in the distance component of target k direction;

3) sorting position in population according to antibody, calculate the distance component d of each antibody in target k direction _k, obtain the affinity of i antibody on k target apart from d _ik;

4) each antibody is sued for peace in the distance component value of all target directions, obtain the crowding distance of each antibody;

The correlation computations formula is as follows,

$d_i=\underset{k=1}{\overset{M}{\mathrm{\&Sigma;}}}\frac{d_{\mathrm{ik}}}{D_k}$ (formula 6)

d _ik＝f _(i+1，k)-f _(i-1，k)

D _k＝δ+max(f _k)-min(f _k)

In formula, d _ibe the crowding distance of i antibody, M is the quantity of optimization aim, d _ikspan is 0-1; f _{(i+1, k)}the desired value of k optimization aim of i+1 antibody, max (f _k) be maximal value in k optimization aim of all antibody of antibody population, min (f _k) be minimum value in k optimization aim of all antibody of antibody population, δ is a normal number.

Characteristics of the present invention: have generally accurately, science, robotization, intelligent characteristics, various land resource space layout optimization methods with respect to widespread use in Practical Project at present, the problem that the present invention mainly solves has: (1) takes full advantage of the advantage of multiple goal Artificial Immune Algorithm aspect optimization problem, and the multiple goal Artificial Immune Algorithm is introduced to solving of land resource space layout optimization problem; (2), according to the characteristics of land resource space layout optimization problem, antibody coding scheme, optimization aim function, constraint condition system, initializer and antibody variation operator towards land resource space layout optimization problem have been designed.

The accompanying drawing explanation

The basic flow sheet that Fig. 1 is multiple goal artificial immunity optimized algorithm (NICA) in prior art.

The antibody coding schematic diagram towards land resource space layout optimization problem that Fig. 2 is the embodiment of the present invention.

The ultimate principle of antibody spatial variability in the land resource space layout optimization that Fig. 3 is the embodiment of the present invention.

The basic flow sheet of the antibody spatial variability operator that Fig. 4 is the embodiment of the present invention.

The Pareto front-end view of the multiple-objection optimization that Fig. 5 is the embodiment of the present invention.

Embodiment

Below in conjunction with drawings and Examples explanation technical solution of the present invention.

For the purpose of understanding, Theory of Interpretation basis at first:

The space layout optimization problem of land resource is the multiple goal combinatorial optimization problem of a class Problem with Some Constrained Conditions from seeing in essence.The multiple goal artificial immune system is as a kind of outstanding multi-objective optimization algorithm, obtained successful Application in a lot of fields.Existing research shows, in the multiple-objection optimization field, artificial immune system in convergence of algorithm speed, keep the diversity etc. of population many-sided more outstanding than intelligent optimization algorithms such as genetic algorithm, ant colony optimization algorithms.In the multiple goal artificial immune system, the NICA algorithm is proved to be one of a kind of outstanding multiple goal artificial immunity optimized algorithm.Therefore, the NICA algorithm that the present invention be take in the multiple goal artificial immune system is basis, builds the multiple goal artificial immunity intelligent optimization model towards the land resource space layout.Wherein, the basic procedure of NICA algorithm is shown in accompanying drawing 1, and the related notion of multiple-objection optimization is as follows:

(1) Pareto is dominant: for given two decision vector x1 and x2, and corresponding object vector y1 and y2, if a certain desired value in decision vector y1 must be better than y2, and all the other desired values of y1 are not inferior to all the other desired values of y2, claim that x1 is that Pareto is dominant to x2, x1 is non-domination in other words.The mathematical description that Pareto is dominant can be with reference to pertinent literature.(formula (2) Pareto optimum solution: for disaggregation X and one of them feasible solution x1, if do not exist any x2 to be dominant to x1Pareto, claim that x1 is Pareto optimum solution or non-domination solution (Non-Dominated).

(3) Pareto optimal solution set: the set that all Pareto optimum solutions form is called the Pareto optimal solution set.

In accompanying drawing 1, the ultimate principle of NICA algorithm is as follows:

(1), for any artificial immunity antibody population, the principle that can be dominant according to Pareto, be divided into Pareto disaggregation and non-Pareto disaggregation by it.

(2) generate initial population in random mode, and the desired value vector of each antibody in population is calculated;

(3) principle be dominant according to Pareto, be divided into Pareto disaggregation and non-Pareto disaggregation, arranges disaggregation Ab _?<Dom}with non-domination disaggregation Ab _?<Non};

(4) clone (Cloning), separate the Pareto of population each antibody of concentrating, copies c doubly, forms new interim population A b _?<New};

(5) variation (Mutation), made a variation to each antibody in new population, obtains new population Ab _?<New}*;

(6) estimate (Evaluation), the desired value vector of each antibody in the new population after variation is calculated;

(7) Pareto sequence, by new population Ab _?<New}merge with original Pareto disaggregation, and carry out the Pareto comparison that is dominant according to the desired value vector of each antibody, obtain new Pareto disaggregation, i.e. non-domination disaggregation Ab _?<Non}and replace original Pareto disaggregation *.

(8) population upgrades, and obtains non-domination disaggregation Ab _?<Non}*.Early stage at algorithm iteration, because the Pareto disaggregation scale obtained is too small, can take random mode to generate the solution of some to increasing the scale of Pareto disaggregation; In the algorithm iteration later stage, because Pareto disaggregation scale is excessive, while affecting the algorithm execution efficiency, according to the crowding distance between antibody, delete too intensive antibody.The specific implementation of relevant NICA algorithm, can consult pertinent literature, and it will not go into details in the present invention.

The land resource space layout optimization method based on the multiple goal artificial immune system of the embodiment of the present invention, the specific implementation process is as follows:

One, obtain planning region at year in base period (present situation) appraisal of land suitability figure as a result, and adopt raster file to be stored.Wherein, the appraisal of land suitability result has comprised each piece soil for different suitability degree, and its span is (1-100), and it is more suitable that the suitability value shows more greatly.For example, certain piece soil is 50 to the suitability of ploughing, and to the suitability in forest land, is 80, shows that this soil is more suitable for using as field.Grid size is arranged according to the range size of planning region, and desirable 10 meters 10 meters usually, 20 meters * 20 meters, 25 meters * 25 meters, 50 meters * 50 meters, several specifications such as 100 meters * 100 meters.When planning region area large (at county level), can adopt 50 meters * 50 meters, the grid of 100 meters * 100 meters or larger specification, when the planning region area hour (village, town level) with 10 meters 10 meters, 20 meters * 20 meters, the grid of 25 meters * 25 meters sizes.The suitability score value obtains according to the appraisal of land suitability working result, can adopt existing techniques in realizing or directly import the result of estimating in advance, and the present invention does not do and repeats.

Two, coding, by integer coding, be mapped as a kind of land resource space layout scheme the artificial antibody of artificial immune system, and stored with the two-dimensional array of a big or small M*N in computing machine.Wherein, M means line number, and N means columns.A grid in planning region represents an actual plot, a gene in corresponding artificial antibody, an array element in corresponding two-dimensional array.Wherein, the characteristic information of the locus (ranks number) that gene has comprised corresponding plot and ground class (comprising the land used type information that suitability score value and current plot to all ground class configure), the specific coding principle is shown in accompanying drawing 2; Therefore, the array element in the two-dimensional array of mentioning in the present invention, plot, grid cell, picture dot and gene indication are identical concept.

In Fig. 2, land use pattern is taked the mode of integer coding, and arable land, field, forest land, meadow, city-building land used and Land Use of Rural Residential Area are corresponded to respectively to integer 1,2,3,4,5,6.The land used type information of original state is empty, follow-uply through the antibody population initialization, is configured.Two-dimensional grid on space divide and accordingly the class coding formed a soil and utilized the space layout scheme, coding wherein is some states of cell, being likely present status of land utilization, is likely random definite, is likely also to determine according to the size of suitability score value.; Soil utilizes the space layout scheme all can adopt aforesaid way to be described arbitrarily, and is stored in the mode of two-dimensional array.In two-dimensional array, the line number of array element (Row) and row number (Col) have been determined this array element correspondence position spatially; A1, the attribute fields such as A2, for recording the suitability score value of corresponding each the land used type in this plot, extract the corresponding grid of figure as a result from appraisal of land suitability.For example, A1 can be used for storing the corresponding suitability score value of ploughing in this plot, and A2 is used for storing the suitability score value in corresponding field, this plot, the like.

Three, objective function setting, the objective function of land resource space layout optimization is mainly designed according to the relative theory of landscape ecology, mainly comprises two optimization aim:

(1) maximum suitability degree (Maxmazation of the overall land suitability)

The First Principles of land resource space layout is the space distribution of all kinds of lands used of arrangement that will suit measures to local conditions.Carrying out the optimization of land resource space layout while optimizing, must improve as far as possible the suitability that in zone, soil utilizes, to accomplish " to the greatest extent its with ". its mathematical description is:

$S=\mathrm{Max}\left({\mathrm{\&Sigma;}}_{i=1}^N{\mathrm{Suit}}_i\right)---\left(1\right)$

In formula, N means the sum of grid cell in planning region, Suit _iit is the suitability score value of the ground class that i grid cell is current configured to it.

(2) maximum compactness (Maxmazation of the spatial compatness)

According to Landscape Ecology Principle, in land resource space layout scheme, on space, grid picture dot adjacent and that the land used type is identical has formed patch.The size of patch, shape flow and produce material impact the protection of local area ecological systemic-function, matter and energy.For example, for the arrangement of the agricultural such as arable land, field, forest land or ecological land, should form as far as possible the large patch of concentrating in flakes, on space, realize to a certain degree gather to facilitate management, cost-saving, also be conducive to farming land protection simultaneously; To embody the principle of " centralization and decentralization combine " for Land Use of Rural Residential Area, the to a certain degree suitable concentrated construction due to rural area infrastructure on the one hand, and need on the other hand dispersion to a certain degree, to reduce farming distance and Production&Operations Management cost.

Setting up exactly of " maximum compactness " target in the optimization of land resource space layout is optimized for the figure spot size and shape to all kinds of soils on space; make it more to be conducive to the farming land protection, reduce costs of production and operation; improve and produce Discussing Convenience etc., its mathematical description is:

$\mathrm{Comp}=\mathrm{Min}\left({\mathrm{\&Sigma;}}_{i=1}^n{\mathrm{LSI}}_j\right)---\left(2\right)$

In formula, the value that Comp is maximum compactness target, n is the number of planning land region utilization bunch (patch).The theoretical minimum value of Comp is n.The value of Comp is less, shows that the shape of each patch in the space layout scheme is more excellent.LSI _jbe the shape index of j patch, its mathematical description is:

$\mathrm{LSI}=\frac{P}{4\sqrt{A}}---\left(3\right)$

In formula, P means the girth of patch, and A means the area of patch.

Four, constraint condition setting, the land resource quantitative structure that land resource space layout prioritization scheme is corresponding must equal the land resource quantitative structure that regional planning is set, and its mathematical description is:

${\mathrm{\&Sigma;}}_{t=1}^N{a}_t\&times;x_{\mathrm{tl}}=A_l,x_{\mathrm{tl}}\&Element;\left\{\mathrm{0,1}\right\}---\left(4\right)$

In formula, A _lfor planning the area of the l class land used of stipulating in year land resource quantitative structure in zone, N means the number of grid cell in planning region, a _tit is the true area in t grid cell representative plot.X _tlbe a decision variable, when the ground of grid cell class is l, value is 1, otherwise value is 0.

Five, antibody population initialization, establishing the chromosome population scale is AN, in conjunction with land resource quantitative structure project period, takes random mode to produce AN initial antibodies, concrete steps are:

(1) build list Y, the number that the length of Y is the Land in Regional Land use pattern, each element representation needed grid number of class accordingly in Y.

(2) produce empty antibody, the value of each gene of antibody is empty.

(a 3) i grid cell can be labeled as plot i, i.e. gene position i.Travel through each gene position, for current gene position i, the land suitability score value based on picture dot builds wheel disc, adopts the roulette strategy to determine at random its ground class.The principle that wheel disc builds is as follows

$p_l=\frac{f_l}{\underset{l=1}{\overset{L}{\mathrm{\&Sigma;}}}{f}_l}---\left(5\right)$

In formula: p _lfor the selected probability of land used type l, f _lfor the suitability score value of current plot to land used type l, L is the sum of class accordingly in list Y.Plot i is higher to the suitability score value of certain class land used, and the probability of the selected type of the land used as current plot of this land used type is larger.The random random number rd that produces (0-1), the judgement position of rd in wheel disc, and then the ground class L of definite current plot i _i, and by ground class L _ivalue corresponding in list Y deducts 1, if the array element value subtracted after 1 is 0, it is deleted from list Y.

(4) repeating step (3), until list Y is empty, can complete the generation of initial antibodies.

Six, antibody desired value vector calculation, employing formula (1) and formula (2) are calculated the desired value vector of each antibody, carry out the Pareto sequence that is dominant, and obtain Pareto and optimize disaggregation, as the memory antibody population.Specific implementation is prior art, and it will not go into details in the present invention

Seven, clone, according to clone's coefficient C, copy C part by current memory antibody, forms new antibody population;

Eight, variation, travel through new antibody population, and each antibody is implemented to mutation operation, and ultimate principle is shown in accompanying drawing 3, and the operator flow process is shown in accompanying drawing 4.In figure, at first with random mode determine the plot (gene position) that will be made a variation (as in figure 1.), as the variation position, and then analyze soil, periphery plot, current plot and utilize situation (being the neighborhood location mode) and compactness, suitability information, determine the target land used type that current plot will change; Take random mode according to target land used type and spatially search for the plot (in Fig. 4 2.) of the target land used type that the land used type will change as current plot; Land used type (in Fig. 4,3., type 6 and 2 exchanges) is intercoursed in final two plot.Huge for land resource space layout optimization problem, traditional Stochastic Optimization Algorithms was restrained slow problem, and the present invention is improved the Mutation Strategy of artificial immune system, added the guidance of domain knowledge, and the operator flow process is shown in accompanying drawing 4.Traversal antibody gene position i, establishing gene position sum (picture dot sum) is N, the initial value of i is 0.

The process flow diagram of contrast accompanying drawing 4, the concrete steps of antibody gene variation are as follows:

(1) i=i+1, the ground class of current gene position is L _i.Generate the random number rnd between (0-1), if rnd is less than the aberration rate P of antibody _m, enter next step.Otherwise jump to next gene position, continue to repeat this step.P wherein _marrange by the user as algorithm parameter, usually get 1/N.

(2) obtain the neighborhood picture dot collection P of current gene position, and the ground class comprised in statistics P and corresponding picture dot number thereof.

(3) if only comprise a kind of ground class in neighborhood P, return to step (1), otherwise enter next step.

(4) probability of occurrence of various places class, i.e. various places class L in statistics P _jshared number percent in neighborhood P, be designated as probability P _j.

(5) obtain the suitability score value S of the corresponding plot of current gene position to all types land used _j.

(6) calculate the comprehensive selection probability C of various places class _j=P _j* S _j.

(7), according to the comprehensive selection probability of various places class, adopt roulette strategy (being the roulette wheel method) to select at random a ground class L _cground class after implementing to make a variation as current gene position.

(8) search out a ground class in antibody and be encoded to L _cgene position C, as the gene swapping position, and with current gene position exchange ground class coding, i.e. the ground class of switch i and position C.

(9) if i<N returns to (1) and continues to carry out, otherwise exits circulation, antibody variation has operated.

In above-mentioned steps, step (8) is found a ground class and is encoded to L in antibody _cand can be further divided into the following steps again with the algorithm of the gene position of current gene position exchange ground class coding:

(1) a position y of the random generation in the zone except current gene position in antibody gene.

(2) if the ground class coding L of position y _yfor L _c, enter next step, otherwise return to step (1).

(3) obtain the neighborhood picture dot collection P of position y, and the quantity of interior ground of statistics P class.If only have a kind of ground class in P, return to step (1), otherwise enter next step.

(4) if comprise L in the neighborhood picture dot collection P of position y _iloop termination, using position y as antibody gene exchange position.Otherwise return to step (1) and continue to search the exchange position.

Nine, antibody desired value vector calculation, employing formula (1) and formula (2) are calculated the desired value vector of each antibody.

Ten, population upgrades, the memory antibody population obtained in antibody new population and last iteration cycle is merged, and the population after being combined re-starts the Pareto sequence, obtain new Pareto disaggregation, obtain new memory antibody population, adopt the new memory antibody population this time generated during next iteration.

Separate quantity for limiting maximum Pareto, after carrying out the new Pareto disaggregation of Pareto sequence acquisition, the present invention further proposes: statistics Pareto separates the quantity TP that concentrates Pareto to separate, if TP is greater than default maximum Pareto and separates quantity MP, according to the crowding distance size of each solution, delete TP-MP the Pareto that crowding distance is less and separate, remaining Pareto is deconstructed into new memory antibody population.Wherein, the computing method of antibody crowding distance are as follows:

1) according to the desired value antagonist population of k optimization aim, sorted;

2) sort the most front and antibody most end is set to a positive infinity value in the distance component of target k direction;

3) sorting position in population according to antibody, calculate the distance component d of each antibody in target k direction _k, obtain the affinity of i antibody on k target apart from d _ik;

4) each antibody is sued for peace in the distance component value of all target directions, obtain the crowding distance of each antibody.

Computing formula is as follows:

$d_i=\underset{k=1}{\overset{M}{\mathrm{\&Sigma;}}}\frac{d_{\mathrm{ik}}}{D_k}$ (formula 6)

d _ik＝f _(i+1，k)-f _(i-1，k)

D _k＝δ+max(f _k)-min(f _k)

In formula, d _ibe the crowding distance of i antibody, M is the quantity of optimization aim, d _ikspan be (0-1); f _{(i-1, k)}the desired value of k optimization aim of i+1 antibody, max (f _k) be maximal value in k optimization aim of all antibody of antibody population, min (f _k) be minimum value in k optimization aim of all antibody of antibody population.For D between the distance regions that prevents k target _k(is 0, and it is a very little normal number that δ is set.Two optimization aim are arranged in the present invention, so M=2, the value of k is 1 or 2.

11, end condition judgement, if algorithm current iteration number of times reaches the maximum iteration time G of user preset, program stops, and decoding obtains the Pareto prioritization scheme, otherwise returns to step 7.By programmed decision-making person according to actual needs, therefrom select the final planning alternatives of one or more conducts.

Parameter arranges suggestion: whether the key parameter setting of algorithm has rationally also determined the performance of algorithm and the accuracy of operation result, and the parameter that relates to parameter and suggestion arranges as follows:

Parameter	Parametric description	The suggestion value
			AN	The initial population scale	Suggestion value 50-200
C	Clone's multiple	Suggestion value 3-6
			MP	Maximum pareto disaggregation scale	The suggestion value, 50-200
Pm	The variation probability	Suggestion value				1/N, N is the picture dot sum
			G	Iterations	Suggestion span (1～10) * N

For the purpose of description effect, choose certain county's land resource space layout optimization problem as case, need to carry out intelligent optimization to the space layout of its arable land, field, forest land, meadow, city-building land used and Land Use of Rural Residential Area.Wherein, grid picture dot size is 50 * 50 meters, distributing unit rationally is 860,000, in computing machine, coding is realized the optimization method that the present invention designs, it is 100 that the initial population scale is set, and maximum non-domination disaggregation scale is 64, and clone's multiple is 4, aberration rate is 900,000/, evolutionary generation is 90 all ages.After carrying out optimization, obtain the Pareto front end and see Fig. 5.Wherein, in Fig. 5, leg-of-mutton point is suitability value corresponding to this Area Land Resources space layout present situation and compactness value.As shown in Figure 5, the Pareto disaggregation after optimization in each Pareto suitability target and the compactness target of separating all be better than present situation.In real work, can consider suitability target and the compactness target of prioritization scheme, selection scheme A is as final programme

Only below that the specific embodiment of the invention case is described, but not in order to limit practical range of the present invention.All equivalent deformations, replacement or modification that those of ordinary skill in the art complete under the spirit indicated without prejudice to the present invention and principle, still be included in the scope of the claims in the present invention.

Claims (7)

1. the land resource space layout optimization method based on the multiple goal artificial immune system, is characterized in that, comprises the following steps:

Step 1, by the planning region appraisal of land suitability as a result figure adopt raster file to be stored;

Step 2, coding, encode a kind of land resource space layout scheme by two-dimensional integer, be mapped as the antibody of artificial immune system; Comprise with a grid cell in planning region and represent an actual plot, a gene position in corresponding artificial antibody, the characteristic information of locus ranks that each gene position has comprised corresponding plot number and ground class, described characteristic information comprises the land used type information to the configuration of the suitability score value of all ground class and current plot, the suitability score value according to appraisal of land suitability as a result the raster file of figure obtain;

Step 3, objective function setting, comprise two optimization aim, is respectively maximum suitability degree S and maximum compactness Comp,

$S=\mathrm{Max}\left({\mathrm{\&Sigma;}}_{i=1}^N{\mathrm{Suit}}_i\right)---\left(1\right)$

In formula, N means the sum of grid cell in planning region, Suit _iit is the suitability score value of the ground class that i grid cell is current configured to it;

$\mathrm{Comp}=\mathrm{Min}\left({\mathrm{\&Sigma;}}_{i=1}^n{\mathrm{LSI}}_j\right)---\left(2\right)$

In formula, the value that Comp is maximum compactness target, the number that n is patch in planning region, on space, grid cell adjacent and that the land used type is identical forms patch, LSI _jit is the shape index of j patch;

Step 4, constraint condition setting, comprise that setting land resource quantitative structure corresponding to land resource space layout scheme must equal the land resource quantitative structure that regional planning is set;

Step 5, antibody population initialization, establishing the chromosome population scale is AN, in conjunction with land resource quantitative structure project period, takes random mode to produce AN initial antibodies;

Step 6, antibody desired value vector calculation, comprise that employing formula (1) and formula (2) calculate the desired value vector of each antibody, and carry out the Pareto sequence, obtains Pareto and optimize disaggregation, as initial memory antibody population;

Step 7, clone, comprise current memory antibody copied to C part according to default clone's coefficient C, forms new antibody population;

Step 8, variation, comprise the antibody population that traversal is new, and each antibody is implemented to mutation operation;

Step 9, antibody desired value vector calculation, employing formula (1) and formula (2) are calculated the desired value vector of each antibody.

Step 10, population are upgraded, and antibody new population and original memory antibody population are merged, and the population after being combined re-start the Pareto sequence, obtain new Pareto disaggregation, obtain new memory antibody population;

The judgement of step 11, end condition, reach default maximum iteration time G if calculate the current iteration number of times, termination of iterations, and decoding obtains the Pareto prioritization scheme.

2. the land resource space layout optimization method based on the multiple goal artificial immune system according to claim 1, it is characterized in that: the shape index LSI of patch obtains as follows,

$\mathrm{LSI}=\frac{P}{4\sqrt{A}}---\left(3\right)$

In formula, P means the girth of patch, and A means the area of patch.

3. the land resource space layout optimization method based on the multiple goal artificial immune system according to claim 1, it is characterized in that: the performing step of step 5 is as follows,

Step 5.1, build list Y, the length of Y be the number of class, each element representation needed grid cell number of class accordingly in Y;

Step 5.2, produce empty antibody, and the value of each gene position of antibody is empty;

Step 5.3, travel through each gene position, and for current gene position, land suitability score value based on plot builds wheel disc, adopts the roulette strategy to determine at random its ground class, the principle that wheel disc builds as shown in the formula,

$p_l=\frac{f_l}{\underset{l=1}{\overset{L}{\mathrm{\&Sigma;}}}{f}_l}---\left(5\right)$

In formula, p _lfor the selected probability of ground class l, f _lfor the current plot suitability score value of class l over the ground, L is the sum of class accordingly in list Y; A random random number rd that value is 0-1, the judgement position of rd in wheel disc, and then the ground class L in definite current plot of producing _i, and by ground class L _ivalue corresponding in list Y deducts 1, if the element value subtracted after 1 is 0, this element is deleted from list Y;

Step 5.4, repeating step 5.3, until list Y is empty, complete the generation of initial antibodies.

4. the land resource space layout optimization method based on the multiple goal artificial immune system according to claim 1, it is characterized in that: establish the gene position i of traversal antibody, take grid cell as picture dot, establish gene position and add up to N, the initial value of i is 0; The performing step of step 8 is as follows,

Step 8.1, i=i+1, the ground class of current gene position is L _i, generate the random number rnd of value between 0-1, if rnd is less than default antibody variation rate P _m, enter next step; Otherwise continue to repeat this step and jump to next gene position;

Step 8.2, obtain the neighborhood picture dot collection P of current gene position, and add up ground class and the corresponding picture dot number comprised in P;

Step 8.3, if only comprise a kind of ground class in neighborhood P, return to step 8.1, otherwise enter next step;

Step 8.4, various places class L in statistics P _jprobability of occurrence, be designated as probability P _j;

Step 8.5, obtain the suitability score value S of the corresponding plot of current gene position to all types land used _j;

Step 8.6, the comprehensive selection probability C of calculating various places class _j=P _j* S _j;

Step 8.7, according to the comprehensive selection probability of various places class, adopt ground class L of the random selection of roulette strategy _cground class after implementing to make a variation as current gene position;

Step 8.8 searches out a ground class and is encoded to L in antibody _cgene position C, with current gene position exchange ground class coding;

Step 8.9, if i<N returns to step 8.1 and continues to carry out, otherwise exits circulation, antibody variation has operated.

5. the land resource space layout optimization method based on the multiple goal artificial immune system according to claim 4, it is characterized in that: step 8.8 comprises the following steps,

Step 8.1.1, a position y of the random generation in the zone in antibody except current gene position;

Step 8.1.2, if the ground class L of position y _yfor L _c, enter next step, otherwise return to step 8.1.1;

Step 8.1.3, obtain the neighborhood picture dot collection P of position y, and add up the quantity of ground class in P, if only have a kind of ground class in P, returns to step 8.1.1, otherwise enter next step;

Step 8.1.4, if comprise L in the neighborhood picture dot of y _iloop termination, using position y as antibody gene exchange position, otherwise return to step 8.1.1, continues to search the exchange position.

6. according to claim 1 or 2 or 3 or the 4 or 5 described land resource space layout optimization methods based on the multiple goal artificial immune system, it is characterized in that: in step 5, for new Pareto disaggregation, the quantity TP that statistics Pareto separates, if TP is greater than default maximum Pareto and separates quantity MP,, according to the crowding distance size of each solution, delete TP-MP the Pareto that crowding distance is less and separate, remaining Pareto is deconstructed into new memory antibody population.

7. the land resource space layout optimization method based on the multiple goal artificial immune system according to claim 6, it is characterized in that: the crowding distance computing method that each Pareto separates corresponding antibodies are as follows:

1) according to the desired value antagonist population of k optimization aim, sorted;

2) sort the most front and antibody most end is set to a positive infinity value in the distance component of target k direction;

3) sorting position in population according to antibody, calculate the distance component d of each antibody in target k direction _k, obtain the affinity of i antibody on k target apart from d _ik;

4) each antibody is sued for peace in the distance component value of all target directions, obtain the crowding distance of each antibody;

The correlation computations formula is as follows,

$d_i=\underset{k=1}{\overset{M}{\mathrm{\&Sigma;}}}\frac{d_{\mathrm{ik}}}{D_k}$ (formula 6)

d _ik＝f _(i+1，k)-f _(i-1，k)

D _k＝δ+max(f _k)-min(f _k)

In formula, d _ibe the crowding distance of i antibody, M is the quantity of optimization aim, d _ikspan is 0-1; f _{(i+1, k)}the desired value of k optimization aim of i+1 antibody, max (f _k) be maximal value in k optimization aim of all antibody of antibody population, min (f _k) be minimum value in k optimization aim of all antibody of antibody population, δ is a normal number.

Images