组搜索优化算法附matlab代码

function [X FX]=gso( fobj, U, L, options ) % GSO group search optimizator % Written by Gustavo de Souza (2011) % Based on the algorithm proposed in the article: % "Group Search Optimizer: An Optimization Algorithm Inspired by Animal Searching Behavior" % by He, S., Wu, Q.H. and Saunders, J.R. % % Parameters: % @fobj: objective function % U: vector (1 x n) with upper limits, where n is the number of dimensions % L: vector (1 x n) with lower limits, where n is the number of dimensions % options: options structure for algorithms % popsize: population size % elitesize: proportion of population in elite group % nproducers: number of producers % nscroungers: number of scroungers (in % of population) % a: range constant % tmax: maximum pursuit angle % amax: maximum turning angle % lmax: maximum pursuit distance % niterations: maximum number of iterations % X=[]; FX=[]; %% verifica se dimensoes de U e L sãiguais if sum(size(U) ~= size(L)) disp('Incompatible sizes: size(U) != size(L)\n'); return; end; %% numero de dimensoes dim=length(U); %% read options gsoptions.popsize = 100; gsoptions.elitesize = 10; gsoptions.nproducers = 1; gsoptions.nscroungers = 0.8; gsoptions.a = round(sqrt(dim+1)); gsoptions.tmax = pi/(gsoptions.a)^2; gsoptions.amax = gsoptions.tmax/2; gsoptions.niterations = 5000; gsoptions.error = 0.0001; gsoptions.stall = 50; gsoptions.verbose = 1; if exist('options','var') gsoptions = options; end; %% verbose level verbose = gsoptions.verbose; %% Moving average window size windowsize=gsoptions.stall; %% Janela (rank) FitWindow=9999; %% limite minimo para busca for i=1:gsoptions.popsize Xmin(i,:)=ones(1,dim).*L; end; %% limite maximo para busca for i=1:gsoptions.popsize Xmax(i,:)=ones(1,dim).*U; end; if ~(exist('options','var')) gsoptions.lmax = sqrt(sum(Xmax(1,:)-Xmin(1,:))^2); end; % population size popsize = gsoptions.popsize; % numero de membros da elite elitesize=gsoptions.elitesize; % numero de produtores NumProducers=gsoptions.nproducers; % numero de oportunistas NumScroungers=ceil(gsoptions.nscroungers*(popsize-NumProducers)); % numero de exploradores NumRangers=popsize-NumProducers-NumScroungers; % fator multiplicador para os rangers (a) arange = gsoptions.a; % maximum pursuit angle tmax=gsoptions.tmax; % maximum turnung angle amax=gsoptions.amax; % maximum pursuit distance lmax=gsoptions.lmax; % limit search methos limitspace = gsoptions.limitspace; % erro maximo erro=gsoptions.error; % numero maximo de iteracoes niterations=gsoptions.niterations; % Geraç da populaç inicial phi=ones(popsize,dim-1)*pi/4; X=Xmin+rand(popsize,dim).*(Xmax-Xmin); % Avaliar os membros Fit = evaluatePopulation(X,popsize,fobj); % Total de movimentos dos produtores e rangers total_moves_p=0; total_moves_r=0; R=[]; P=[]; P_cont = zeros(1,popsize); phi_0 = zeros(popsize,dim-1); if (verbose>0) printf('Iterations: %d\n',niterations); end; for iteration=1:niterations % ponteiros para os tipos de membros R_old = R; P_old = P; [P, S, R] = selectMembers( NumProducers, NumScroungers, Fit ); % Perform producing for np=1:NumProducers n=P(np); if (verbose>0) if find(n==R_old) printf("Goldmine!\n"); end; end; if find(n==P_old) P_cont(n) = P_cont(n) + 1; end; r1=randn(1); r2=rand(1,dim-1); D =sph2car(phi(n,:)); Dr=sph2car(phi(n,:)+r2*tmax/2); Dl=sph2car(phi(n,:)-r2*tmax/2); % Escolhe 3 pontos dentro do hipercubo Xz = X(n,:) + r1*lmax*D; Xr = X(n,:) + r1*lmax*Dr; Xl = X(n,:) + r1*lmax*Dl; % % limit the search space % Vmax=find(Xz>Xmax(n,:)); Xz(Vmax)=X(n,Vmax); % Vmin=find(Xz<Xmin(n,:)); Xz(Vmin)=X(n,Vmin); % % Vmax=find(Xr>Xmax(n,:)); Xr(Vmax)=X(n,Vmax); % Vmin=find(Xr<Xmin(n,:)); Xr(Vmin)=X(n,Vmin); % % Vmax=find(Xl>Xmax(n,:)); Xl(Vmax)=X(n,Vmax); % Vmin=find(Xl<Xmin(n,:)); Xl(Vmin)=X(n,Vmin); % Calcula o fitness fXz=fobj(Xz); fXr=fobj(Xr); fXl=fobj(Xl); % Compara os pontos e escolhe o movimento Xnovo = X(n,:); fXnovo = Fit(n); phinovo = phi(n,:); if fXz < fXnovo Xnovo = Xz; fXnovo = fXz; phinovo = phi(n,:); end if fXr < fXnovo Xnovo = Xr; fXnovo = fXr; phinovo = phi(n,:)+r2*tmax/2; end if fXl < fXnovo Xnovo = Xl; fXnovo = fXl; phinovo = phi(n,:)-r2*tmax/2; end if fXnovo < Fit(n) X(n,:)=Xnovo; phi(n,:)=phinovo; Fit(n)=fXnovo; ir=-1; total_moves_p = total_moves_p + 1; P_cont(n) = 0; phi_0(n,:) = phi(n,:); else % if the producer can not find a better area after a iterations, % it will turn its head back to zero degree. if P_cont(n) > 5 if (verbose>0) printf('FOCUS!!!\n'); end; phi(n,:) = phi_0(n,:); P_cont(n) = 0; else r2=rand(1,dim-1); phi(n,:)=phi(n,:)+r2*amax; end end end %% Perform scrounging for ns=1:length(S) n = S(ns); % Each scrounger randomly selects a producer to move towards np = P(unidrnd(length(P))); r2=rand(1,dim-1); r3=rand(1,dim); X(n,:) = X(n,:)+r3.*(X(np,:)-X(n,:)); phi(n,:) = phi(n,:)+r2*amax; Fit(n) = fobj(X(n,:)); end %% perform ranging for nr=1:length(R) n=R(nr); r1=randn(1); r2=rand(1,dim-1); phinovo=phi(n,:)+r2*amax; li = arange*r1; D=sph2car(phinovo); Xnovo = X(n,:)+li*D*lmax; % limit the search space Vmax=find(Xnovo>Xmax(n,:)); Xnovo(Vmax)=X(n,Vmax); Vmin=